Long before anyone trained a model on customer data, the marketing data business worked out the terms for pooling it. Catalog retailers in the 1990s joined data cooperatives that combined their transaction histories so each member could find new buyers, and the arrangements held together because the rules were blunt.

A retailer contributed data to receive modeled prospects; no member got a rival's customer file; and a company that stopped contributing stopped receiving. Card fraud consortia run on similar give-to-get terms. Nobody would call these systems elegant, but every participant understood the trade.

Software vendors building AI features are now running into the same question, usually without the rules. Traditional software sat on top of a customer's data and operated on it. AI products learn from data, and on a multi-tenant platform, that learning rarely stays within a single account.

One company's records can sharpen the enrichment, lead scoring, and recommendations delivered to thousands of others. A vendor can improve its product for every customer by relying on the work of a few, and most contracts say nothing about whether that trade is fair. The customer still owns its data in any legal sense.

Ownership alone doesn't answer who benefits.

Users push back on sharing hard-earned data

HubSpot learned how touchy the question has become when it announced plans to pool enrichment data across customer accounts, including business contact details, employer information, and email deliverability signals, to improve prospect records platform-wide.

Every account was in by default. Within days, the company dropped the plan and promised that any future enrichment feature built on customer data would be opt-in. HubSpot had stated clearly that core CRM records would never be shared without permission, so privacy in the usual sense wasn't what upset the sales and RevOps teams, who pushed back. Their concern was harder to answer. Years of work spent cleaning and enriching their own contact data were about to improve the prospecting tools available to every other customer, competitors included, and nobody had asked what they would get for it. The product terms that allowed the change had been updated nearly two years before most customers heard about them.

AI creates accidental data companies

Plenty of software companies adding AI tools have never had to think this way. They are becoming data companies one model at a time, often without deciding to. Whether a customer's information feeds a shared model is often determined by a product team working toward a launch date rather than by anyone who owns the commercial relationship, and the answer is expressed in terms few buyers read.

McKinsey's 2026 survey on AI trust found that only about 30% of organizations have reached a mature level in AI strategy, governance, and controls. Pooled training data tends to fall between legal, product, and sales, with no single team feeling responsible for explaining it.

The need for strong exit clauses

Buyers will force the issue long before regulators do. Security certifications and uptime guarantees became standard contract language over the past decade because enough customers asked for them that vendors stopped arguing, and AI data terms are heading down the same road. Procurement teams are starting to ask whether their data trains or enriches models used by other customers, whether participation is opt-in, and whether it can be switched off later without losing core features. The more demanding buyers want to know what a contributing customer receives that a non-contributor doesn't.

Exit terms will be the hardest part to write. A customer can cancel a contract and demand its records back, but data that has already shaped a production model can't be cleanly pulled out again. Retraining is expensive, and removing one customer's influence from a live model remains an unsolved problem for most vendors. A vendor can honestly say it deleted every record a customer ever uploaded and still be running a model those records helped build. Contracts written for storage assumed that deleting data ended its use. Contracts written for AI will have to say plainly what persists after a customer leaves, and whether the customer agreed to it when it signed.

Opt-in consent answers part of the question. A checkbox tells a vendor it may use the data, but it says nothing about the return, and a customer who opts in for nothing in particular has made a poor bargain, even if it's a legal one. The old cooperatives signed up members because members could see exactly what they were getting.

Vendors that explain their data economics before customers ask will have an easier time at every renewal. The explanation belongs in the product and in the contract, and it should cover which customer data feeds shared models, what those models produce, and what contributors get back. Security questionnaires were once a formality, until buyers made them a condition of the sale. Pooled data terms are on the same schedule, and vendors that write clear ones now will set the standard that the rest of the market will follow.

I was trying to buy something online and I did what most people do now: I let an AI overview point me to it. It sent me straight to a product page, confidently. And guess what happened when I clicked it? The product was sold out.

It happened again a few weeks later, a different item, this time discontinued. And the more I thought about it, the more it bothered me, not because the AI was wrong, exactly, but because of why it was wrong. Somewhere, a company had a page that hadn't been updated and an AI system had no way of knowing that. It just read what was there and repeated it confidently, to me.

That's a small, annoying version of a much bigger problem companies are about to run into everywhere.

Most companies have spent the last decade getting serious about data. There's a data warehouse, a Chief Data Officer, a governance committee that meets monthly, and dashboards running the quarterly business review. On paper, the discipline is all in place.

Nobody applies that same discipline to the information the business actually runs on. A policy gets updated in one system and quietly stays wrong in three others. Someone leaves, and the page they updated quarterly just stops getting maintained. A product goes out of stock, or gets discontinued, and the page stays live and indexed for months. None of this looks like negligence. It looks like a normal Tuesday.

It used to be fine, because people compensated for it. A salesperson knew the pricing page was stale and quoted the real number anyway. A new hire figured out within a month which internal wiki page to actually trust. But it is becoming evident that the workaround only worked because a human was in the loop, reading the room, catching the gap before it reached someone who mattered.

AI doesn't have a loop to be in. It doesn't know the product's sold out unless something tells it directly. It doesn't pick up on the office joke that nobody's touched that page since last spring. All of that context lived in people's heads, not in the document, and a model can only read the document.

So the gap doesn't get quietly caught anymore. It gets published straight to the customer.

Where the accountability actually breaks down

Ask a CEO who owns the company's information and you'll get an org chart back. Marketing owns the website. Product owns the specs. Legal owns policy language. And if the company operates in more than one region, each one has probably built its own version of everything above.

Every team is doing its job, but collectively, the company ends up saying five different things about itself. For a long time, that only mattered if a human happened to stumble across the contradiction. Now an AI system finds it for you, instantly, and hands it to a customer as a fact.

A recent consumer survey found that nearly half of shoppers double-check an AI answer before acting on it, and when what they find contradicts it, the business loses a customer it technically just showed up for. That's exactly what happened to me that I noted earlier. The AI didn't lie. It just repeated something that was no longer true, confidently enough that I believed it until I clicked through.

That's the real issue. Information ownership rarely breaks down at the point something gets published. It breaks down after. The person who wrote a page isn't usually the person who finds out, months later, that it's gone stale. Publishing has an owner, but that information staying correct usually doesn't.

A few questions worth asking

If your company sells any product online, or publishes anything customers rely on, it's worth thinking who actually owns catching it when this information goes stale, not just publishing it in the first place.

And it's worth checking, directly, what an AI system currently says about your company or your product to someone who's never talked to you before. Not what you assume it says, but what it actually says that could affect your business.

Nobody planned for AI to be the thing that exposes every quiet inconsistency a company has been living with for years. But that's exactly what it's doing and companies need to take action. The ones that will get ahead of it aren't the ones with the fanciest AI strategy, but the ones who decide who's responsible for making sure what's out there is still true.

We've reviewed and ranked the best website builders.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Ask a room of executives about agentic AI and within about ninety seconds someone will ask how many agents they ought to be running. It has become the default measure of progress: a number you can put on a slide, benchmark against a competitor, and report upwards.

It is also the wrong question.

Across the 200-plus workshops my team and I have run with enterprises in Europe and the Middle East, the organizations getting real value from AI tools are not the ones with the longest agent inventory. They are the ones who never framed the problem that way to begin with. They started by identifying a business problem or point of friction and worked backwards. Everyone else started from the technology and went looking for somewhere to put it.

There are four questions I hear constantly that are quietly steering good companies into Pilot Purgatory. Here they are, and here is what I would ask instead.

Don't ask how many agents. Ask what outcome you are improving.

Much of the last two years of enterprise AI has been about helping people work faster. Copilots that answer questions, assistants that summarize the meeting you have just sat through, tools that get you past the blank page. These are genuinely useful and I would encourage everyone to switch them on.

But speed is not the same as value, and agent count is not the same as either. No enterprise ever created value by accumulating agents. Value shows up in real business outcomes such as fewer payroll errors, faster shift scheduling, lower supplier sourcing costs, better win rates, cash collected sooner.

That is the difference between an agent and an agentic application. An agent help with an individual task. An agentic application coordinates teams of specialized agents to deliver a business outcome. If you cannot name the KPI an agent is moving, you do not have a use case. You have a demo.

Don't ask whether AI can recommend. Ask whether it can finish the job.

The most persistent confusion I encounter is between advice and execution. Plenty of systems can now sense a situation, spot the pattern and tell you what should happen next. That is real progress. It also leaves almost all of the work exactly where it was.

Take sourcing. An assistant can flag that a category is overdue for renegotiation and suggest you run an event. Someone still has to create that event, chase the responses, score them against commercial criteria, route the approvals correctly, update the contract and shepherd the whole thing to a close.

An agentic application does that work, under policy, with human oversight and approval where required, and with a record of what it did and why.

The test is simple. At the end of the interaction, is the process further forward, or do you just have a better-informed human facing the same queue?

Don't ask which model tops the leaderboard. Ask whether it understands your business.

Benchmark scores dominate the AI conversation and matter far less in the enterprise than most leaders assume. Raw model intelligence is rarely the binding constraint.

A frontier model understands procurement as a discipline. It does not understand your supplier list, your delegation-of-authority matrix, your contractual obligations, your current inventory position, or the three exceptions your finance director signed off in March. Without that, sophistication amounts to a well-argued guess.

Enterprise context is the differentiator: trusted enterprise data and business context, including the policies, approvals, transaction history, organizational structure and live process state that already sit inside your core systems. The closer agentic capabilities are to the systems where business processes, data, security, and approvals already live, the more effectively agents can understand context and act within establish controls.

Don't ask for one system that does everything. Ask how the specialists coordinate.

There is a persistent hope that enterprise AI will eventually resolve into a single, omniscient digital assistant. Businesses do not work that way, so their AI will not either.

Organizations are ecosystems of specialists: finance, HR, supply chain, service, sales, each with its own expertise, obligations and objectives. The equivalent in AI is not one agent that knows everything, but coordinated teams of agents that know their own domain and work towards a shared goal.

Hiring is a good illustration. One agent surfaces and screens candidates. Another prepares onboarding. A third confirms the budget is genuinely available. A fourth keeps the hiring manager's approvals moving. Individually, each is modestly useful. Together they move a business process from vacancy to first day. The value sits in the coordination, not the components.

Systems of outcomes

What connects all four questions is a shift in what enterprise software is actually for. For decades our applications recorded what had happened. We are now entering the era of systems of outcomes: software that reads the state of the business, reasons over it, and acts to keep work moving towards a defined objective.

That shift demands different questions from leadership. Not how many agents, but which outcome. Not can it advise, but can it execute. Not how clever is the model, but how well does it know us. Not one brain, but one team.

Get the questions right and the technology decisions become considerably easier. Get them wrong and you will assemble an impressive collection of agents, tools, and models that nobody can tie to a number that matters.

We've featured the best AI website builder.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Digital sovereignty today is less about where data sits and more about who controls it. It is the ability for organizations to govern their data, infrastructure and operations, while retaining the flexibility needed to innovate.

Sovereignty is about having true control over your data and knowing where it lives, how it is regulated and exactly how it can be accessed at all times. Organizations can approach sovereignty through balancing regulatory requirements, operational resilience and innovation.

Different industries face vastly different pressures when it comes to data sovereignty. Banks, insurers, and healthcare providers must protect sensitive data while staying agile, while tech companies or logistics firms often require a greater level of flexibility.

In plain terms, data sovereignty is impacted by how data is governed by the laws of the country or region it is physically stored and/or accessed in. Data ownership and control are crucial, but current conversations about sovereignty have become a catch-all for resilience and security – which goes beyond data residency alone. It is never just about the physical location of the data.

The main challenge organizations face now is how to determine the right level of sovereignty for their regulatory, operational and innovation specific needs.

Sovereignty across sectors

When it comes to regulation, different industries face entirely different regulatory and operational pressures, with no single universal mandate applying across all sectors.

In highly regulated industries like healthcare, government and financial services, organizations are required to meet strict local data controls to comply with sector-specific regulations, one example being NIS Regulations within the EU. Organizations working with sensitive data must ensure they are protecting personal and financial data while they remain operationally flexible.

For example, a global bank holds highly sensitive customer data like account details and transaction history and would therefore need to operate under dense and overlapping regulatory frameworks. Both customers and regulators expect strict control over the data, requiring the organization to uphold a specific and tightly governed sovereignty framework in its foundations.

Organizations in less regulated industries like tech or retail can often leverage global cloud provider infrastructure more freely, without needing to carefully balance performance or innovation.

Sectors that don’t require high levels of sovereignty have to ensure they do not overdo the concept as it often leads to slower development and wasted resources.

Finding your place on the spectrum

Think of sovereignty as a dial which your organization uses to find your place on the spectrum. The dial moves depending on how much control you need, over what data and under which jurisdiction, to find the exact level for your organization and potentially per project or service. This method lets you examine your business in accordance to the regulatory requirements for your sector and location.

There are three main questions you need to ask to guide your decisions. Ask what your actual regulatory obligations are and where operational resilience requires local control vs where it demands global flexibility. Then think about where, in your business, does innovation depend on the free movement of data.

Once you have answered the above questions, you’ll know what level of sovereignty applies to your organization and how and when it might be subject to change. It is important to remember that your sovereignty decision must be revisited as your business scales, enters new markets and jurisdictions and as regulation evolves in your sector.

Why your database is a sovereign decision

Your sovereignty decision is essentially ensuring you understand the importance of security, compliance, accountability and explainability when it comes to your organization's internal business controls and regulatory requirements.

In light of this, database infrastructure is taking on an elevated role and supporting enterprises across industries who choose and tailor the level of sovereignty they require and that’s just right for them.

Modern database software platforms can support enterprises across the spectrum, with strict geo-fenced deployments with data localization and flexible multi-cloud database architectures for more global businesses. To become a strategic decision, sovereignty has to be considered at the data layer of your infrastructure.

Technical capabilities like end-to-end encryption, strict audit and access controls and scalable support across the globe are what often matter in a decision about sovereignty. The right model of infrastructure gives organizations the choice to tailor their level of control to their actual needs.

Sovereignty is a strategy

Sovereignty must be viewed as a spectrum that an organisation can choose from. It cannot be treated as a linear standard across industries and organizations.

Enterprises who treat sovereignty as a deliberate strategy are the ones getting sovereignty right. This outlook forces organizations to be precise about the exact thing they are protecting, at what cost and why they are protecting it.

Before asking if your organization is sovereign, think about whether you are sovereign in the ways that actually matter for your business and sector.

We've reviewed and ranked the best cloud storage.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Transparency duties under Article 50 have begun to apply, even as the EU AI Omnibus implementation for many high-risk systems has moved to December 2027 and, for certain AI-enabled products, to August 2028. For organizations building and deploying AI in Europe, the message is clear: regulators extended the timeline on complexity, but not on transparency.

While some organizations may view the extension as welcome breathing room, the time would be far better used preparing for the future of AI. To do that, we need to frame the challenge around creating the conditions for long-term innovation, rather than around compliance deadlines.

As AI becomes embedded in products, services and decisions that affect people's lives, organizations need confidence in the systems they are deploying, and users need confidence in the organizations deploying them. That work cannot begin in 2027. It needs to start now.

Compliance is achieved at a point in time; governance is built as a capability. A deadline can be moved, but an institutional capability cannot simply be created at short notice.

Start with the AI already in use

The starting point for every organization should be an honest assessment of whether it has a reliable picture of the tools, outputs and interactions that Article 50 covers. As we have seen in recent years, AI systems can quickly become more deeply embedded in business over time as employees explore new tools and suppliers innovate and ship new updates.

This natural expansion is a normal part of adopting a new technology, but it can result in AI becoming a core part of business processes before the proper governance is in place. Governance is most effective when it shapes those decisions from the beginning.

Attempting to retrofit governance after AI has become embedded is significantly more difficult, more expensive and ultimately less effective, and the resulting loss of trust can be more costly than the upfront work required to prevent it.

Extend the controls already in place

The good news is that many of the disciplines required by the legislation should already be familiar. They build on existing practices around privacy, security, risk management, accountability and operational governance, and much of the challenge is around applying established governance disciplines to a rapidly evolving technology landscape.

Organizations that have taken data protection regulation seriously, and already understand impact assessments and effective governance processes, are well positioned to meet many of these requirements – there is considerable overlap.

The task is to extend them for AI, not replace them. A privacy impact assessment can identify data risks, but it cannot on its own establish whether a model is dependable in a particular context, whether someone can meaningfully question its output or whether an automated process disadvantages a group.

Those questions should be built into existing workflows, particularly when AI is bought rather than built. Before a capability is embedded in software or a critical process, organizations need clarity on data treatment, service and model changes, incident response, support for transparency duties and the evidence a provider can supply. Even where the underlying model comes from a third party, organizations still need to govern how it is selected, configured and used.

The sovereignty question

Alongside the AI Act, the other crucial question impacting the way we think about governance in Europe is sovereignty. As the debate gains momentum across Europe, the practical issues raised are becoming clearer. Where is data stored? Which providers do we depend on? How resilient are our technology choices? What happens if commercial, legal or geopolitical circumstances change?

It is worth being precise about what sovereignty can realistically mean for enterprises. Full compute sovereignty, including control over chips, infrastructure and supply chains, is far beyond the reach of most organizations. The more practical goal is compute resilience: understanding where critical dependencies sit, ensuring no single vendor, cloud provider or model becomes a point of leverage, and retaining control over the layers of the AI stack that remain within the organization's influence.

Open source can support this by providing greater transparency and control over code, but it does not remove infrastructure dependency on its own. An organization may use an open model and still rely on external compute or cloud capacity to operate it.

Organizations should adopt a sovereign-by-design approach to innovation. Sovereignty should not be considered after technology decisions have been made, it needs to be embedded into decisions on data architecture, vendor selection and risk management from the outset, with resilience, not full independence, as the realistic target.

It is also important to note that not every dataset, process or workload carries the same level of risk. Organizations should take a tiered approach, applying the strongest sovereignty requirements where data sensitivity, regulatory exposure or operational importance demand it.

These are not abstract concerns. Viewing AI governance and sovereignty as interconnected challenges creates an opportunity to build a more coherent approach. This is not a case for technological isolation or for stepping back from global technology ecosystems, the aim is narrower than that. It is about ensuring organizations retain meaningful freedom of action in an increasingly complex and uncertain environment.

The real opportunity

The debate about innovation, regulation and sovereignty will continue. Europe will keep refining its approach to AI, and organizations will keep adapting to new technologies, new providers and new risks. The more important question is whether organizations use this window to build the capabilities they will need long term, rather than treating each regulatory milestone as a separate deadline.

Governance should not be viewed as a compliance exercise that slows innovation. It is an institutional capability that enables organizations to innovate sustainably, adopt new technologies with confidence, and remain resilient as markets, regulations and geopolitical conditions evolve.

The ultimate goal is innovation built on trust, and we have the ability to deliver that regardless of regulatory deadlines. The real test is not whether an organization built every part of the system it relies on, but whether it understands its dependencies well enough to change course if circumstances demand it.

Store your data in the cloud with the best cloud storage service.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The race for AI leadership has largely been framed as a contest for the best chips, models and talent up until now. But behind every algorithm, model, query, and breakthrough is the energy that powers it all.

As AI moves from R&D into living rooms, workplaces, and factories, access to reliable electricity and the speed at which data centers can connect to the grid (AKA “time to power”) is becoming the defining factor determining where data centers (or AI factories) get built and who gets to reap the economic benefits.

For investors, it is no longer enough for a country to promise abundant electricity in the future.

Power must be accessible, predictable, competitively priced, and available on a timeline that matches the speed of the AI market.

A connection delayed by several years can mean missing an entire technology cycle.

Three factors that will shape where AI investment lands

Data center operators make investment decisions based on three fundamentals: regulatory environment, availability of energy and the price of that energy.

Increasingly, however, investors are seeking confidence: confidence that permits will be secured and grid connections delivered on schedule, and power will remain affordable.

China and the US dominate global capacity but are taking different paths. China is investing around US$295 billion in a nationwide AI computing network, alongside energy infrastructure and renewables. In the US, growing electricity demand and evolving grid-connection timelines are increasing interest in on-site generation and storage.

Europe presents a more fragmented picture. Some markets offer available capacity and attractive investment conditions; others face grid queues and complex permitting. In spite of this, Europe has many of the right ingredients. Its mix of nuclear power and renewables, combined with established infrastructure and deep industrial expertise, should make it a natural home for the next generation of AI infrastructure.

Europe may have missed much of the first wave of AI data center build out, but it still has the opportunity to capture the economic benefits of the next one. With targeted investment and smarter energy policy, countries can still reshape the map of AI infrastructure investment.

There are three levers the world should pull now to capitalize the powerful impact AI can have.

1. In the AI economy, speed is power

Energy systems around the world are struggling to accommodate rapid growth in electricity demand. Many of today’s grids weren’t built to manage such a diverse mix of energy sources alongside increasingly intense and concentrated power demand. As such, an estimated 2,500GW of generation capacity is waiting in connection queues worldwide.

In Europe’s data center hubs, often referred to as FLAP-D (Frankfurt, London, Amsterdam, Paris, and Dublin), data centers can wait up to a decade for a grid connection, while a facility may take just four years to build.

Governments and grid operators must accelerate permitting and connection processes. There needs to be more transparency around expected grid connection timelines, supported by delivery milestones and regular reporting so developers and investors can plan with greater certainty. It also means moving from “first come, first served” to “first ready, first connected”, prioritizing viable, grid-supportive projects over speculative applications.

UK efforts to remove these “zombie projects” demonstrate how queue reform can shorten waiting times and strengthen investor confidence. This helped shrink waiting times which were up to 15-years for some projects.

2. Flexibility is the fastest source of new capacity

Grid modernization is essential, but major upgrades could take up to 15 years. While wider grid expansion takes place, flexibility can free up existing capacity more quickly.

In many regions, the challenge is not simply how much electricity is generated, but whether it is available where and when it is needed. Energy-management technologies give operators the visibility and control to optimize energy use across buildings, transport and industry.

They can shift non-critical demand, such as heating, cooling and EV charging, away from peak periods. Battery storage complements this by storing lower-cost or surplus electricity for use when demand is highest. As these capabilities become more widely connected and coordinated, they could unlock greater grid capacity, reliability and flexibility at scale.

Data centers themselves can also become active grid assets through intelligent energy management. If European data centers offered just 30 hours of flexibility per year, the grid capacity available to them could more than double. The technologies needed to achieve this already exist, including digital energy-management systems, battery storage and microgrids.

Compass Data centers offers an early example of what this can look like in practice.

Schneider Electric has already delivered around 150 prefabricated modular data center solutions for Compass, while connected, condition-based maintenance has helped reduce manual on-site interventions by 40% and operating expenditure by 20%.

Alongside microgrid controls being deployed at its Phoenix campus, this shows how data centers can scale more predictably while gaining the visibility and control to manage power dynamically when the grid is under pressure.

Countries now need the right incentives and market mechanisms to reward flexibility and make participation commercially viable.

3. Make flexibility the path to cheapest power

Technology can make energy flexible, but economics will determine whether that flexibility scales. In regions like France, Germany, Italy and the Nordics where grid flexibility exists, households and businesses can be rewarded for using power when renewables are plentiful and reducing demand at peak times.

Flexibility in data centers could mean lower tariffs or faster staged connections in exchange for battery storage, automated demand response and participation in flexibility markets. Real-time monitoring and automated controls can manage these adjustments without compromising performance.

The grid’s largest untapped reserve may not be another power station, but the millions of connected assets already hiding in plain sight.

4. The real race to power AI

AI infrastructure will go where power is available, affordable and resilient. A recent report by Reuters found that European data center developers are moving away from major cities to where energy is cheaper and connections are quicker.

Countries unable to provide those conditions risk losing more than data centers. They could also lose the investment, jobs, industrial ecosystems and innovation that develop around them.

The countries that want to capitalize on the AI boom should be all-in on electricity. That means expanding clean generation, strengthening and digitalizing grids, improving regulatory clarity and unlocking the flexibility inside homes, buildings, transport and industry.

The future of AI will be shaped by data centers. But it will be won where power is abundant, resilient, and affordable. In the AI era, access to electricity becomes access to economic advantage.

We've reviewed and ranked the best web hosting services.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The proliferation of AI tools in business is redefining the need for data security and trust. As autonomous, decision-making AI agents move from pilot projects into core business workflows, they are expanding the enterprise attack surface just as fast as they are being adopted.

Gartner predicts that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, a jump from less than 5% in 2025. Gartner’s advice to leaders of software organizations is to set their agentic AI strategy or “risk being outpaced by the competition”.

Security is changing. What was once a “simple” focus on encryption and perimeter defense is no longer enough on its own: the rise of AI-driven risk has transformed it into a multi-dimensional challenge built on algorithmic trust, model integrity, and synthetic data exposure. Trust boundaries are shifting, and most companies aren’t yet equipped to handle it.

Yet AI has already been proven unsafe by its very makers, with the Hugging Face breach showing it was a rogue AI agent's fault. The unknown future risk of autonomous agents capable of “emergent” behavior poses huge issues for maintaining security. But this is just one risk of many faced by enterprise cyber security leaders.

In fact, the statistics of cyber security investment highlight that business is vastly underprepared for new threats. While the estimated annual cost of cybercrime is $10.5 trillion globally and predicted to reach $15.63 trillion by 2029, this eclipses Gartner’s information security spend prediction of $244 billion in 2026.

Machine data is exacerbating the data quality issue with garbage in, garbage out. This calls for technology and trust architectures to be redesigned, with internet security embedded into the data lifecycle. Security leaders will need to make considered investments in AI-safe security, transparency, and federated governance to protect their enterprise ecosystems.

The biggest security risks we’re underestimating right now

With many threats now AI-driven, it’s unlikely that a single enterprise has protection against all the necessary risks, which is why it’s incumbent on business leaders to prioritize next-level data security.

Firstly, polymorphic, AI-generated attacks are hitting bigger and faster, beyond any company’s human ability to counter these attacks. Deepfake is one such AI-powered theft and extortion technique that can be generated in minutes, costing a mere $1,000 (if that) to create a malicious bot that can hack into your system and go after your business’ crown jewel data.

Using these methods, hackers can affect AI models by manipulating decision-making itself instead of through direct attacks on the IT infrastructure. Shadow AI and exposure to synthetic data are also growing concerns as employees adopt unsanctioned or non-firewalled AI tools, models and workflows to complete mundane tasks and reporting.

Next is the surge in agentic adoption without control of the risk, especially third-party risk, which is huge. Agentic AI has significantly increased the threat surface, with systems capable of autonomously taking actions. In the rush to implement AI, access is being granted for agents to all kinds of data, even to highly sensitive data, often bypassing traditional enterprise controls. All kinds of blind spots emerge when agents are connected to external systems whose security vulnerabilities aren’t vetted.

Prompt injections are another critical risk, on a similar level to the whole third-party dependency. It's the idea that an organization doesn’t know what is actually making its way into their system. At first appearing benign, it can build up to turn malicious at some point in time. When it hits, the business won't know what hit them.

As if this isn’t enough, we’re counting down to “Q-Day”, when quantum computers become powerful enough to break today’s widely used encryption software standards. This is the stage when attackers will be able to intercept, decrypt, and compromise nearly all global digital communications, financial transactions, and other forms of secure data.

Key considerations for a responsible AI approach

What’s needed is a formal framework for trust-enabled ROI in the age of AI. It needs to be about maintaining data source integrity. In order to trust the data, there are various mechanisms or things that an enterprise needs to take care of such as identity and accountability.

We must know who's acting on the data and have context and control. We must then authorize it contextually, not implicitly and this means constantly validating behavior and intent. This is all while staying compliant against or changing the regulatory landscape and maintaining human oversight across the lifecycle.

Responsible AI requires an overall shaping of an approach that considers the impact on the entire value chain in the ecosystem. Only this ensures that the value is delivered to the people who are putting into the system or that are impacted by the system. The following are essential elements to move this forward:

Accountable humans

The concept of accountability is key. There’s a tendency to say if something bad happens, you blame the machine, you blame the algorithm, you blame the system. There need to be accountable humans that take responsibility. AI is here for the long term, and will have generational impacts, so if we don't understand that and take accountability for it, we're going to make wrong decisions.

Four dimensions of responsible AI

At a high level, the focus should center around transparency, accountability, fairness, and protection. Some key questions should be asked as part of building a framework for successful outcomes, such as:

• How do you make sure that your AI models and data practices are transparent?
• How do you make sure you drive accountability for AI-driven decisions?
• How do you make sure you're fair to the people affected by your AI systems?
• How do you make sure that you provide protection or security across the entire data lifecycle?

Building a trust premium

Amidst this new risk landscape, trust has become a measurable competitive asset for enterprises. Organizations that are moving their cyber security strategy to the next level and investing early in AI governance, cryptographic modernization and security architecture are building a trust premium. PwC research backs this up: across thousands of CEOs surveyed globally for its 25th Annual Global CEO Survey, customer trust explained 31% of the variance in profit margins and 21% in return on assets - a stronger link to financial performance than company size.

Durable trust requires an approach that is built and maintained across the entire operational life-cycle of data. This full-lifecycle view of trust comprises data source integrity, end-to-end data quality, identity and accountability, behavioural monitoring and intent validation, regulatory compliance and effective governance, applying human oversight across the AI lifecycle.

Taking a methodical approach to protection

The changing frontiers of data security demand a fresh approach, which involves building resilient data infrastructures and best practices that can stop threats in their tracks. Future successful enterprise security will depend as much on governing autonomous systems as well as defending networks.

The right approach is to take a step back, think holistically, considering all the building blocks that need to be put in place, and then be more methodical. A methodical approach protects the business, its customers, and it protects a society so that we can all progress.

Doing things right in the right way will have multiple cascading benefits, including guaranteed value for customers and the longevity of the business, both of which links to trust and loyalty.

We've Reviewed, Rated, and Ranked the Best Firewall Software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Elvis Presley has been dead for nearly 50 years, but apparently even the 'King of Rock and Roll' can't escape the AI industry's appetite for training data. The owners of 274 Elvis songs made famous by Presley have sued Anthropic, accusing the company behind Claude of using his music without permission to build its AI models, and are seeking tens of millions of dollars in damages.

Anthropic disputes the allegations and says it intends to defend the case. AI companies have long argued that training models on copyrighted works can qualify as fair use, an issue courts are still working through. There's an important legal difference between learning patterns from existing material and reproducing protected expression, but that distinction has become a remarkably convenient place for the AI industry to hide its enormous dependence on other people's creativity.

It's hard to sympathize with the companies on the receiving end of these sorts of lawsuits, even when the plaintiffs are not without resources of their own. Unfortunately, these disputes are becoming all too normal. AI-powered music is becoming more widespread, if not necessarily popular, and we've reached a point where discovering that an AI company allegedly used hundreds of copyrighted songs without permission barely qualifies as surprising news.

The only unusual part is that this time the person whose music is at the center of the dispute happens to be Elvis Presley.

A little less uncompensated song use

The lawsuit makes an accusation that goes beyond simply using copyrighted material for training. According to the complaint, Claude could generate material that closely resembled existing Elvis songs, even after Anthropic introduced safeguards intended to prevent infringement. Requests to recreate songs for modern audiences allegedly produced versions retaining recognizable elements of the originals.

An AI model learning about music isn't automatically infringing copyright, and copyright law doesn't give anyone ownership of an entire musical style. But if a commercial AI system reproduces protected lyrics or creates unauthorized adaptations, it's hard to see why the company behind it should get a special exemption simply because the copying happens through complicated software.

AI companies have been trying out various arguments in court, but a lot of it feels like an attempt to claim that AI development is so important that the usual rules should somehow bend around it. Training requires enormous amounts of information, and licensing everything would be expensive or inconvenient, but that's a standard headache for companies whose products rely on copyrighted material.

TV shows and movies license music, whether an original track or just the right to record a new version for their project. No company is allowed to simply grab hold of copyrighted material simply because clearing it would slow down development. AI companies should have to take copyright seriously too, even if that makes their business models more expensive.

It's also particularly galling that songs people spent years creating and popularizing can become raw material for commercial technology without the rights holders having any meaningful say. Access to a work and permission to exploit it are separate questions, regardless of how sophisticated the software becomes. Ignoring those rules is why artists who passed away decades ago seem to produce new music whose earnings don't go to their estate.

AI music turns the volume up

Elvis is hardly the first musician to become part of this fight. Major music publishers have already sued Anthropic over allegations involving copyrighted lyrics, while AI music generator Suno has faced legal challenges from record labels. The Elvis lawsuit stands out because it focuses on the catalog associated with one of the most recognizable performers in history.

That fame might help draw attention to a problem that has affected considerably less powerful creators for years. Independent artists may struggle to establish whether their work was used to train a model. If someone with Elvis Presley's commercial legacy needs a federal lawsuit to challenge alleged unauthorized copying, ordinary musicians' prospects look bleak.

Of course, the companies controlling Elvis's catalog aren't struggling artists themselves, but that doesn't make the underlying complaint any less important. The principle shouldn't depend on whether the copyright owner is wealthy or famous.

I'm also not convinced that every AI company needs unrestricted access to practically everything ever written or recorded to produce useful technology. Claude is an impressive assistant, but its ability to answer questions doesn't suddenly make songwriters' copyright interests irrelevant. If AI developers want to build profitable products using creative works, they should be prepared to explain where that material came from and why they were entitled to use it.

The courts may decide in Anthropic's favor due to fair use, but even that ruling wouldn't automatically mean AI companies have permission to go wild on every song ever recorded. The best path forward won't be as simple as making it always allowed or illegal to use copyrighted material for AI training, but the companies building these systems shouldn't rely on their ambitions as a good enough reason to ignore rights holders.

The AI industry has become very good at explaining how much it needs human creativity, but it's time it became equally good at respecting the people responsible for it.

Well, it’s finally happened: Apple has got with the times and finally made a laptop with a touchscreen. However, while the new MacBook Pros with PROLED screens look like fantastic devices, I think Apple has made a mistake by putting its first laptop touchscreen on the more expensive MacBook Pro, rather than the more mainstream MacBook Air… or even the MacBook Neo.

The 14-inch MacBook Pro with a touchscreen starts at $2,899 / £2,899 / AU$4,599, and the 16-inch model starts at $3,399 / £3,399 / AU$5,499. That’s a big ask for most people, especially these days.

Of course, MacBook Pros, as the name suggests, are aimed at professionals, and thanks to their high-end components, they should really be seen as investments that allow you to speed up your workflow.

Interestingly the new touchscreen MacBook Pros come with Apple’s M5 Pro and M5 Max chips. If you want Apple’s latest M6 chip, you’ll need to go for the non-touchscreen 14-inch MacBook Pro that was announced a few days earlier. Perhaps confusingly, the M6 is less powerful than the M5 Pro and M5 Max, and so far there are no hints that Apple is preparing to release M6 Pro and M6 Max chips.

My point is that by putting its first-ever laptop touchscreen on its most powerful — and expensive — MacBooks, Apple is severely limiting the potential audience for the technology. Had it instead put a touchscreen on the more affordable MacBook Air, it could have had a real hit on its hands.

Who’s it for?

As someone who has used plenty of Windows laptops and Chromebooks with touchscreens in the past, I know how useful they can be for certain tasks, and I often miss them when using a MacBook.

Being able to scroll down a website with a finger, turn a page in an ebook, or quickly tap a menu item feels natural and intuitive, and I’m sure it will on the new MacBooks. My daughter, who grew up with touchscreens on everything, would also instinctively poke an often sticky, grubby (kids, you’ve got to love them!) finger at a MacBook screen and be surprised when nothing happened.

These are all pretty simple tasks and gestures. What I've never used touchscreens on laptops for, however, is for complex tasks and tools, such as recording and editing music in Ableton Live, or editing videos in Adobe Premiere Pro.

Instead, for things that needed precision or complexity, nothing beat the good old combo of keyboard and touchpad (or even better, keyboard and a mouse), as well as other external peripherals such as MIDI controllers and drawing tablets. Which leads me to wonder why Apple picked the MacBook Pro, a device that’s supposedly designed for exactly those type of tasks, to showcase its first laptop touchscreen.

If you’re dropping $3,000+ on a touchscreen laptop to mostly just to use the touchscreen for browsing websites, then you’re either super-rich or you should get help with your budgeting.

The 16-inch MacBook Pro with M5 Pro and no touchscreen starts at $2,999 / £2,999 / AU$4,799. As the MacBook Pro with the PROLED XDR touchscreen with the same specs costs $400 / £400 / AU$700 more, that’s quite a premium for the touchscreen.

Yes, I know the PROLED XDR screen offers better image quality as well thanks to its tandem OLED tech (check out our explainer on why this is a big deal), but if you don’t need that additional image quality, which includes better contrast, higher brightness levels, and support for the Adobe RGB color gamut, and are just interested in the touchscreen, then essentially yeah, you’re paying $400 extra for it.

If you’re a professional, especially one in the creative industry, you’ll need to think long and hard if that investment is worth it. Meanwhile, the likes of the MacBook Air and MacBook Neo aren’t really used for complex creative tasks, and instead used for casual day-to-day use, where touchscreens are much more useful.

Granted, adding an extra $400 / £400 / AU$700 to the price of a MacBook Neo or MacBook Air wouldn’t be great, especially as they’ve just had a price bump earlier this year, but Apple wouldn’t need to include the OLED tech and advanced image-quality options included in the MacBook Pro. A touchscreen MacBook Neo for $899, say, could be a real hit, especially if Apple was to also keep the non-touchscreen model for people who still want the ultimate affordable way to get into the Mac ecosystem.

This isn’t just an Apple problem, of course. The new laptops powered by Nvidia’s RTX Spark SoC, which are aimed at the same creative professionals and heavy local AI users as the MacBook Pro (with similarly high price tags), come with touchscreens. I’ve spent a decent amount of time with each one at various events, and I found that the touchscreen was the least-used part of the laptop — in fact, I often forgot those laptops had one. Because there aren’t cheaper, non-touch, models, though I didn’t resent its inclusion.

As I said, I do get the appeal of touchscreen laptops, but I find the technology is best used in 2-in-1 devices where the keyboard can be flipped behind the screen (or removed entirely) so that it can be used as a tablet-like device.

In an ideal world, then, Apple would have put its touchscreen on an entirely new 2-in-1 MacBook, but I can’t see that happening any time soon. Apple sort of has a device like that — the powerful iPad Pro, which can be paired with a Magic Keyboard that turns it into a touchscreen laptop. The problem there, however, is that despite coming with Apple’s powerful M-series chips, the same you’ll find in MacBooks, the iPad Pro is limited to the iPadOS operating system and mobile apps.

Put macOS on that bad boy, however, and allow us to run full desktop applications on it, and hey presto: Apple has a touchscreen MacBook that utilizes the tech in a way that’s actually useful. Until then, I’ll stick with my standard MacBook Pro, and just try to remember not to prod it with my sausage fingers too often.

There's no shortage of enthusiasm for artificial intelligence. Organizations across every sector are investing heavily, experimenting, launching proof of concepts, and finding ways to turn the latest developments into real business value.

However, there's a growing disconnect between experimentation and execution. For all the discussions of AI transforming the way we work, many AI projects never progress beyond the pilot stage. It's a familiar story. The proof of concept delivers promising results, the business case adds up, and everything looks like it's heading in the right direction.

Yet somewhere along the way the project hits a bump on the road. This can look like costs beginning to rise, confidence starting to waver, integration with existing legacy systems proving more complicated than expected, and approvals taking longer than anticipated.

What started out as an exciting project slowly loses its momentum. As a result, the AI initiative ends up stuck in limbo. The reason is that running a successful AI pilot and successful AI implementation are two very different things. Transitioning from a controlled environment with governed variables to a production environment brings a new set of challenges: security, governance, culture, skills, integration, and readiness.

Unfortunately, this challenge is not uncommon. Research shows that 71% of organizations remain in the pilot or proof-of-concept phase of AI transformation, highlighting just how difficult it is to turn early promise into a business reality.

This limbo is what many organizations are now experiencing: the AI implementation gap. It's the space between proving that AI can work and deploying it at scale in a way that delivers measurable value. For today's technology leaders, bridging that gap has become one of the defining challenges of AI adoption. The pressure to move beyond experimentation is growing, but doing so requires more than just increasing investment or pushing strong proof of concepts into production.

The real challenge is creating the conditions in which AI can move from an isolated trial into a sustainable part of the wider business. That requires establishing the right technical and organizational foundations, preparing people and processes, and finding the right use cases. To deliver real value, organizations need to think less about speed and more about creating something secure, scalable, and consistent.

It isn’t a technology challenge, it’s a business one

When AI projects stall, it's seldom because of the technology itself. More often, organizations run into strategic, governance, and operational challenges that prevent encouraging pilots from becoming production-ready solutions.

One of the biggest barriers is analysis paralysis. The majority of organizations have no shortage of AI ideas, but they typically lack a clear way of understanding which use cases will deliver the greatest value. This means resources become spread across multiple proof-of-concepts, each showing promise but none receiving the attention needed to reach production. The number of pilots continues to soar while the number of live deployments stagnates.

However, some organizations make the opposite mistake by designing for scale before they've checked the concept works. Major investment is made in infrastructure, architecture, and implementation before a solution has been tested in real-world conditions. If the project then needs to change direction, the amount of investment already committed can make it tricky to adapt. As a result, organizations can end up locked into an approach that no longer delivers the value they envisioned.

Governance is another common stumbling block. For organizations operating across EMEA, navigating regulations such as the EU AI Act, GDPR, and DORA introduces understandable caution. The consequences of non-compliance can be significant, and uncertainty around an evolving regulation often slows decision-making. Many AI initiatives don't stall because the business lacks ambition, but because leaders want greater confidence that they're deploying the technology in the right way: compliantly and responsibly.

Lastly, organizations often underestimate what it will cost to move from a successful pilot to widespread deployment. AI projects that appear affordable during the proof-of-concept phase can become much more expensive once they're scaled, especially if they haven't been properly stress-tested.

That's why due diligence is critical. Yet more than half (56%) of organizations don't carry out a Total Cost of Ownership assessment before deciding where to place significant workloads. Without a clear understanding of the long-term investment required, it's easy for costs to accumulate and confidence in the project to dwindle.

Balancing certainty and speed

Closing the Implementation Gap isn't about moving faster. It's about moving with confidence and know-how.

Time and time again, organizations find themselves at one of two extremes: waiting until every unknown has been resolved before making any moves or rushing ahead with AI projects before they've been properly validated. Neither approach is sustainable. Those seeing the greatest success are the organizations finding a balance between certainty and speed - moving quickly where the evidence supports it, while building in the validation and governance needed to reduce risk.

That starts with validating ideas early. Before major investment is made, organizations need to ask the technical, commercial, and governance questions that could prevent a project from ever reaching production. It's far better to discover early that a use case won't deliver than to invest months of budget and time before uncovering the same conclusion.

Rapid prototyping also plays a crucial role. Instead of debating whether an idea might work in theory, businesses should aim to create something tangible that users can test within days. Seeing a concept in action changes the conversation from speculation to evidence, giving stakeholders the confidence to make informed decisions about where to invest next.

Equally important is maintaining transparency throughout the delivery process. AI projects shouldn't feel like a black box. Defined ownership, clear checkpoints, and agreed assessment criteria help everyone to understand where a project stands, whether it’s a success, and what comes next. Projects rarely stall because the technology isn't capable - they stall because critical decisions weren't made early enough or because expectations weren't aligned from the outset.

Bridging the implementation gap

The implementation gap isn't going to disappear overnight. With 71% of organizations still stuck between pilot and production, the challenge is clearly not a lack of willingness, technical ability or ideas. In most cases, they have all three.

Instead, it's the ability to turn experiments into solutions that actually work in the real world. That means treating the transition as a business issue, not simply a technical one. Organizations need an approach that balances speed with governance, validates value before investment is made, and gives stakeholders the confidence to scale. It also requires leaders to be confident enough to stop projects in their tracks if they're not going to deliver real value.

The organizations taking meaningful steps forward won't be those adding more pilots into the mix. They'll be those that have strong processes and capabilities in place to push the right pilots ahead. Fundamentally, closing the AI implementation gap means creating a path from pilot to production that can be followed again and again.

We've featured the best AI tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Technology innovation should always be driven by one core objective – to improve people’s lives. Unleashing human imagination in new and exciting ways, whether that’s solving complex problems or making everyday life easier.

We’re at a fascinating moment in time, where smartphones have become the de facto platform, but in terms of innovation they have reached a bit of a plateau. New smartphones have largely incremental hardware updates and optimizations without groundbreaking new features.

But then AI came along to shake things up. And that AI, when used well, is making a huge positive impact in people’s lives. AI is best used to complement human abilities – if you look at fields like medical imaging, AI can help to analyze huge data sets and spot patterns for physicians to diagnose fully.

Across tech, AI is having a big impact in how we interact with technology. Most notably transforming smartphones and smart devices. And it’s created a new category where you don’t even need to be holding your phone – AI smart glasses.

The AI smart glasses category has been growing rapidly, with research by Omdia finding global shipments of AI glasses had reached 8.7 million units in 2025, up 322% year-on-year.

It’s interesting to look at the ‘why’. On paper, AI glasses don’t have as much processing power or storage or RAM as a smartphone, so how could they be of use? And it's down to the form factor and AI.

Potential for smart glasses

There’s a lot of potential for smart glasses in the B2B sector. Increased productivity, real-time interactivity or enhanced customer experiences, the potential use cases are huge across many verticals.

One example I love is the potential to change how people engage with cultural subject matter. Imagine a pair of AI glasses as an “audio guide 2.0” that would become a deeply personalized guide, interacting with the knowledge base and curation of the venue – whether it’s a museum, art gallery or historical heritage site.

Visitors can ask specific questions about certain subject matter and get real-time answers as they explore a museum or gallery, making it much more conversational and interactive. Lightweight in form factor, smart glasses can be worn over long periods of time. And in the future, the same tour could generate layers of explanations tailored to different audiences (children, amateurs, experts), without imposing the same experience on everyone. We’re already seeing these concepts being explored by cultural institutions.

This form factor can also blend seamlessly into everyday life, and make tasks easier. How many times have you been somewhere like a concert and witnessed a sea of smartphone screens? The same with navigating around a city. People experiencing life through their device. AI glasses are changing that, allowing you to be more present in the moment without having to live and capture memories through a phone screen.

I actually found myself running late and lost in Rotterdam train station recently. I speak three languages fluently, but if you’ve been to the Netherlands you know it’s hard to guess at some of the words. Safe to say, it was getting stressful. I was wearing AI glasses, and asked it to translate the signs around me – I was on my way quickly.

But with every emerging technology comes new challenges and societal considerations. And there’s a responsibility on manufacturers and innovators to find ways to realize this vision responsibly. For AI smart glasses, there's one crucial piece of the puzzle that will shape their true impact - and that's privacy.

Creating a privacy-first architecture

One element is data privacy. We’re in an era where people’s personal information has been largely commoditized. Where AI smart glasses are concerned, we’re seeing a huge amount of contention about how personal data is gathered by some manufacturers and used to train AI models.

For me, it’s very simple. Our company DNA has always been that privacy should be treated as a core principle, not an afterthought.

But what does privacy-by-design actually look like in practice?

Control of data should ultimately remain with the user. And that’s why it’s crucial to have data stored locally on smart glasses by default – and only transferred off the device when initiated by the user.

And you can safeguard local data even further with military-grade encryption – something that is particularly important for many organizations and will be required when handling internal and customer data. This means that even if someone else tries to take the AI glasses and pair them to another phone, the previous data cannot be accessed.

These concerns over data should also extend to how we interact with LLMs. As AI tools become embedded into our daily lives, the more requests we send to LLMs, the easier it is for them to build a profile of you. And it’s the same for AI smart glasses.

With a privacy-first architecture, third party AI requests are de-identified, so that LLMs cannot build an accurate profile. Policies where user data is not used to train any LLMs should be standard.

There’s also the matter of protecting the privacy of people around you when wearing smart glasses. It’s crucial to have visual indicators, like an LED indicator that lights up during photo or video capture to signal photography or video is being recorded at that moment.

It's obviously not good enough that on some AI glasses, it’s possible to obscure the LED and effectively hide it from view. Any AI glasses worth their salt should use a multi-sensor approach so that photos and video recording should automatically disable if the LED is obstructed or when the glasses are removed.

We collectively have this incredible opportunity - a new technology which combines the versatility of AI with wearables. But AI smart glasses will only be a huge benefit to us all, if privacy is handled correctly.

We've featured the best VR headsets.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

It is now getting pretty difficult to track every time OpenAI and Anthropic models break containment, wander onto the wider web, and cause unintended, high-risk chaos.

The recent wave of rogue AI agent incidents has now prompted US President Donald Trump’s administration to consider asserting more power over AI tools.

These incidents expose a deep vulnerability for the UK. If advanced AI models are acting unpredictably with sensitive data and carrying out hacking campaigns against real people, what happens when that data is hosted, processed, and governed on foreign-owned infrastructure?

For far too long British IT has essentially handed the keys to US tech giants. The CMA’s market investigation into cloud services found that Amazon and Microsoft control 70-90% of the UK cloud computing market. And another US digital powerhouse in Google controls a smaller share.

These hyperscalers often repeat marketing platitudes about "sovereign" regional data centers located on British soil. I am sorry to be the bearer of bad news but the act of storing data in a UK-based data center is not enough to guarantee sovereignty.

In reality, these tech giants are "sovereignty-washing." Why? Because If the controlling entity is subject to foreign legal jurisdiction, such as the US CLOUD Act, foreign authorities can compel access to that data. It is irrelevant where the physical hardware resides. Hosting data on foreign-owned servers is basically just paying rent on digital turf that remains subject to external laws.

As foreign-owned autonomous AI agents process vast amounts of critical enterprise and state data, and increasingly turn adversarial, the decision to build Britain's AI future on foreign, proprietary stacks looks ever more short-sighted. The potential for financial, legal and national security breaches is simply not worth the risk.

It is time for the UK to end its over-reliance on sovereign-washed cloud providers and establish true digital independence by backing British tech.

Paying the "sovereignty tax"

By continuing to rely on foreign tech monopolies British institutions are paying a heavy "Sovereignty Tax.” By that I mean the insidious, hidden and steadily accumulating burden of building critical systems on platforms you cannot fully govern.

Hyperscalers reel in UK organizations using aggressive cloud credits and low upfront pricing. But once they are embedded in these cloud systems, those firms often find it hard to leave them. This can come in the form of data egress fees, proprietary technical debt, outage exposure, restrictive licensing rules and audit blind spots. All of these add to the creeping realization that leaving your current provider would be harder, and more expensive, than you ever anticipated.

This explains why an overwhelming 66% of UK IT leaders say they would consider switching cloud providers to regain control of their data and infrastructure. Yet despite this clear desire to break free, the proportion of UK IT leaders expecting to become even more deeply entrenched in US hyperscaler systems has more than doubled over the past year, rising from 12% to 28%.

This seems all the more bizarre given that a recent study from Proton Research showed that 45% of consumers would actively avoid doing business with companies that store data with US firms. Meanwhile, over half (58%) would happily pay higher fees to switch to a sovereign provider.

This forced entrapment dynamic has undoubtedly created an era of digital feudalism. A small group of foreign tech lords own the underlying digital land, while British businesses and public sector bodies exist merely as serfs… and pay a perpetual tribute for the right to operate. And the real-world operational costs of this reliance are coming home to roost. Over the past year, 39% of UK firms were hit by outages originating from US hyperscalers, with resulting risk exposure costing some organizations over £1 million.

As organizations increasingly deploy more and more compute-intensive AI workloads on these platforms, the technical lock-in will only continue to deepen.

Not only does this drain commercial value but it also means high-skilled engineering jobs, and IP heads across the Atlantic rather than staying in the UK.

Legal and security risks

As autonomous AI usage proliferates, legal and operational risks do so too. AI agents interact with database layers, corporate IP and sensitive citizen data while executing the tasks they are set.

If that data resides in a foreign-controlled ecosystem, UK organizations lose jurisdictional control over it. Without true data sovereignty, external powers effectively hold a "kill switch" over the UK’s critical infrastructure. For example, if a UK NHS trust is reliant on a foreign-hosted AI agent for patient triage, a regulatory dispute in another country could trigger a sudden access shutdown. This could paralyze hospital operations and put sensitive health records in the hands of foreign control.

By centralizing the UK’s national capability into a single foreign-owned monoculture, we are also creating a concentrated point of failure. What happens if the US shifts policy and revokes access, leaving the UK powerless over its own critical infrastructure?

The drone and missile strikes that hit Amazon Web Services facilities in the UAE in March that brought two of the three Availability Zones in the region offline was also clear proof that the cloud is not a borderless, invulnerable utility. For the UK to retain true security, it requires resilient distributed architecture.

British digital independence

It should be noted that global hyperscalers undoubtedly have a place in non-sensitive, low-risk applications. But prioritizing digital sovereignty is not an act of digital isolationism.

The fact that the UK government created a £500 million Sovereign AI initiative this April shows that policymakers do recognize the importance of digital resilience and maintaining control.

But that support for domestic innovation will continue to be undermined if public compute spend continues to be funneled towards multi-year deals with foreign incumbents.

The government’s procurement strategy must focus on actively purchasing from domestic cloud providers if its strategic ambition for creating a UK tech powerhouse is to work.

At the same time, proprietary vendor lock-in thrives in closed-loop ecosystems. The UK needs to treat open-source tech and open architectures as national assets. When you use open standards, moving your AI models and data between public, private, and local clouds is painless. You get actual flexibility instead of being locked into one setup. And, crucially, UK organizations avoid being hit with exorbitant transition fees.

Our European neighbors are already leading the way in regard to embracing true sovereignty. France is dropping Zoom and Office for sovereign, open-source ecosystems, and Germany is using Nextcloud to replace Google Docs.

The UK possesses capable domestic cloud platforms. We have the technical capacity and talent ready to scale. And UK firms are crying out for the powers that be to back homegrown tech; 90% of IT leaders are calling for a more proactive government stance in support of British companies. We just need the procurement bravery to match.

Controlling our destiny

The unpredictable and unprecedented nature of autonomous AI agents means the UK must reassess what it has been doing up to now. Building the future of British tech and economic security on rented land is fundamentally contradictory.

The UK already has companies that own the full tech stack required to power modern enterprise. These companies offer the same core capabilities as hyperscalers without the associated sovereign risk or vendor lock-in.

If the UK wants to prove itself as a global leader in AI the country needs to support its own tech providers.

That means distancing ourselves from sovereign-washed cloud providers and investing in open-source infrastructure. Doing so is the only way to protect our data and reclaim our digital autonomy, before it’s too late.

We've featured the best AI website builder.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Since the launch of Anthropic's Mythos, the debate around the risks that AI-powered tools pose for security teams has steadily ramped up.

Yet, fundamentally, the attack targets aren’t changing: the big shift is one of scale and speed. AI finds more vulnerabilities, giving security teams more to deal with. But at the same time, AI is lowering the technical barrier to exploitation. The result is a situation where the window of time that teams have to respond has drastically reduced.

This has clear implications for how organizations approach exposure management and reinforces the need for solutions that help security and IT teams stay secure amidst this new reality.

Vulnerability detection is accelerating

The initial update from Project Glasswing underlined the power behind the Mythos model that has fueled so much speculation and concern.

In approximately one month, Mythos and around 50 partner organizations reportedly identified more than 10,000 high and critical severity vulnerabilities, with 6,202 estimated high/critical vulnerabilities found across more than 1,000 open-source projects.

There’s no doubt Mythos is hugely consequential and its capabilities mark a significant milestone, but the gap between attackers and defenders has already been widening for some time. Security teams have been facing a growing volume of CVEs; in 2025, some 48,185 CVEs were published, a 20.6% increase from the previous year and Advances in AI tools can reasonably be expected to continue fueling this trend for years to come.

This all has huge consequences for security teams. As Cloud Security Alliance analysis has reported, Mythos-type capabilities will mean the number and frequency of complex, novel attacks will rise dramatically alongside the need to patch and respond to AI-discovered vulnerabilities, exploits and autonomous attacks.

0-day vending machines, no Mythos required

Restricting access to Mythos and other highly capable models is not a practical safeguard. Our research and testing demonstrates the capabilities of general access LLMs in finding and exploiting novel vulnerabilities when being prompted by experienced security practitioners.

One of our security researchers created an automated vulnerability research pipeline, with no human in the loop. On first deployment, this found a remote, multi-stage SQL injection vulnerability sitting on 300,000 WordPress plugins, a process fully automated from discovery to exploitation. The finding, which aligned with an independent discovery by another researcher, was assigned CVE 2026 3985 and publicly disclosed.

And it’s not just the power of AI in finding vulnerabilities that should cause concern. AI-powered attacks can go a step further in automating an entire extortion attempt. An agent can autonomously create and send the exploit, then demand a ransom as we saw when an operator dubbed JadePuffer, exploited a vulnerable Langflow server, harvested credentials, moved laterally and encrypted 1,300 database records in an end-to-end agentic ransomware operation.

Reduce the attack surface as a first line of defense

This is our 'new normal' and LLM models have changed the old assumptions about the speed of vulnerability management. The window that existed between vulnerability discovery, CVEs assigned and defensive action being taken, has fundamentally shifted.

It’s not just about reaction times. Teams also need to take proactive steps to reduce their attack surface as the race to patch can be avoided entirely if assets weren’t unnecessarily exposed in the first place. This is about fighting the battle on two fronts; closing the window between vulnerability detection and remediation and taking pre-emptive action to make the target that attackers aim at smaller in the first place.

These should go hand-in-hand as many teams focus on patching what’s vulnerable, but they also need to question if it should be exposed at all. Attack surface reduction is an area that many organizations may struggle with, but any unnecessarily exposed internet facing asset is now carrying far more risk than ever.

Exposed services, such as leaving a MySQL database or private API documentation exposed to the internet, is an open invitation for attackers and these often don’t even need a CVE to be exploited. This leaves the door open for automated, high-speed attacks.

Always on exposure management

When time to exploit is reduced down to hours, the second key focus area for security teams is reducing the gap between discovery and remediation.

Finding and fixing vulnerabilities requires both context and cadence. Pentesting provides the depth and context but is a ‘point in time’ activity and frequent in-depth investigations can be cost prohibitive for many organizations. Vulnerability scanning provides the cadence but can’t interpret if what it’s found matters for your business environment.

That’s where AI can help and, whilst the traditional pentest isn’t going to disappear right away, we’re at the beginning of an era in which AI offers a new way to support the work that human security analysts and engineers are doing.

Its value lies in instantly alerting teams to new risks, the ability to deploy a pentest on every major release for the fraction of the cost of a manual engagement, agents that can actively investigate and validate findings using the same methods employed by human security experts, and integrating with the tools your teams use to get issues to the right people for fixing, fast.

This way, stretched teams can keep up with the emerging threat landscape. AI is helping teams cross that last mile - validating findings, reducing false positives, and providing necessary context.

These are still relatively early days and the frameworks around the pentest, used by auditors, insurers and compliance bodies, will need to evolve. But the signs are that this new era requires continuous testing as standard practice.

It’s needed because Mythos, and other LLM models, will punish anyone still working to yesterday's timelines. The fundamentals of reducing exposure, validating what's real, and freeing skilled people from the noise haven't changed. What's changed is how little time is left to get them right.

We've featured the best endpoint protection software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The Garmin Fenix 9 Pro is Garmin's latest — and greatest — watch. The Fenix series always occupies a flagship position in Garmin's lineup: while the company does make more expensive watches, most of those are specialist pieces of gear, such as the Tactix for soldiers, the Descent for professional divers, and the Marq series as an ultra-premium option for fashion-forward Garmin users.

The Garmin Fenix series is its top-tier 'generalist' watch, and considering it only comes out once every few years, anticipation is always high. Fortunately, the Fenix 9 Pro lives up to the hype, and is my Fitness Devices of the Year at the TechRadar Product of the Year Awards 2026.

It's a powerhouse of a watch, from construction and performance, and has coped with every test we have thrown at it, from hikes in Chamonix, to comparison tests against the Apple Watch Ultra 4 and an electrical heart rate monitor, the Frontier X2.

Let's start with construction. The Fenix 9 Pro's entire case is now made of titanium, rather than a titanium bezel and fiber-reinforced polymer (that's plastic, to you and me) elsewhere.

It's thinner, and the lack of visible screws (aside from the back plate) make for a refined, sleek aesthetic. The display is sapphire crystal as standard, too, protecting the super-bright AMOLED touchscreen display, with the usual MIP Power Glass solar charging options as an additional extra.

The construction makes for an amazing watch, and these upgrades make the Fenix 9 Pro really feel just that, Pro — more like a Marq. I'm wearing the 43mm version, and it really doesn't feel like an ultra-chunky Fenix. In fact, it feels far more like wearing an upmarket metal dive watch.

This sealed unit design necessitated a redesign in the buttons, according to Garmin's promotional material, in order to make sure it remained waterproof up to 40m deep, but Garmin cracked it and the watch retains Garmin's distinctive five-button design.

It's even revamped the LED light, with an introduction of a third green option to add to its black and red tones. The green light offers improved visibility compared to the red option, without sacrificing your night-eyes.

Software-wise, the UX is familiar to Garmin watch users, as everything is innately customizable, from the widget stack to workout views to watch faces. New features include Garmin Epic, which allows you to string several workouts together into one mega-event, such as multiple ski runs over one long weekend, for example. You can start it from the watch, and view or share it in the app. Rucking also got a useful redesign, while Maps has been updated to provide an angled POV option.

In our performance test, I was expecting accuracy — after all, the Garmin Fenix 9 Pro retains Garmin's Elevate V5 heart rate sensor array from previous years — but I wasn't expecting it to cleave so close to the heart rate monitor and the Apple Watch Ultra 4 (which Apple claims is the most accurate wearable to date) during my test workouts.

Rerouting course directions has proven to be responsive and speedy, while the battery performs as described, lasting for a week, and the watch is comfortable to wear during sleep, workouts and every day.

There wasn't really any doubt that the Garmin Fenix 9 Pro was going to be a good watch, or even a great one. Like the Apple Watch Ultra 4, it builds on everything Garmin has done before to create a worthy successor to the ever-popular Fenix line. When the Fenix 10 comes out in a few years, this one will come down in price just like the Fenix 8 before it, and we'll see far more adopters take it on and have fun with it.

I suppose I could have chosen something like the cheap Google Fitbit Air, the Garmin Cirqa, or the Apple Watch Ultra 4 to be our Fitness Device of the Year. But when I thought about what was going to dominate my limited wrist real estate if I sign up for a race, there was only ever going to be a handful of options, and the 43mm Garmin Fenix 9 Pro was first on the list.

It's smart enough not to look out of place in the office, it's got everything I need for training, it's got all the complications I need from the Connect IQ score, like Spotify and Komoot, and a week-long battery life (sorry Apple) that cinches it for me over the Ultra 4.

In my review of Apple's iPhone Duo, I noted the engineering and build quality, how easy it is to unfold the phone, and also how solid and sure it feels when you close it. And I have opened and closed the flexible, dual-screen phone many, many times.

When a small set of iPhone Duo owners complained that their large inner screens had unexpectedly and disturbingly cracked, Apple did something unusual: it acknowledged the complaints: "We are aware of an issue impacting the inner display on a small number of iPhone Duo units. If users experience issues with their iPhone Duo display, they should contact AppleCare for a replacement.”

That's not Apple saying there's a fundamental flaw or that the issue is found on anything more than a small number of the already popular phones.

I watched with some dismay as people started bandying about the 'crackgate' phrase. Apple has had its share of 'gates' over the years, including 'bendgate' with the iPhone 6 Plus, and Antenna Gate with the iPhone 4. On the latter, Steve Jobs memorably told customers they were holding the phone wrong (he later apologized for the issues).

The notion of '[anything]gate' comes from the Watergate Scandal, a cover-up of a crime involving the then President of the United States, Richard Nixon. Now, whenever people think there's an issue and someone is trying to obfuscate it, they slap a 'gate' on it. I've written before that I hate this.

Leaving aside the term, there is the fact that some people claim the iPhone Duo cracked when they opened it and only shortly after they brought it home. Since we can't rewind those tapes, we have to take them at their word. Still, I think it's fair to wonder if they opened it too roughly, like a strongman trying to tear a phonebook in half. We just don't know.

Inside the iPhone Duo Display

Here's what I do know, though: Apple spent considerable effort building a durable folding iPhone. Seriously, how could it not?

Samsung has been doing this for a decade, and with each iteration of the Samsung Galaxy Z Fold line it has improved the display panel structure and supporting hinge technology. On the Galaxy Z Fold 8 Ultra it actually uses two titanium plates. One relatively rigid plate sits at the bottom of the stack, and is responsible for making the phone open stiffly flat. The other, for durability at the crease, is a super-thin sheet of titanium alloy.

Apple does not talk about where it sourced its iPhone Duo display technology, though most have surmised that it started with Samsung. Even so, the display and hinge architecture of the iPhone Duo is distinct from that the Galaxy Z Fold 8 Ultra.

It's worth digging into these details to understand the iPhone Duo's potential weak points, and because, in many of my conversations about why the Duo displays might have cracked and how it happened, people want to understand what part, exactly, is cracking. The build, as described by Apple on the Surprise and Shine launch day in September, provides some clues.

It's been more than a month since Apple unveiled the iPhone Duo, but I recall Chief Hardware Officer Johny Srouji explaining in stark detail how Apple builds the iPhone Duo display and hinge (complete with images).

According to Srouji, the goal for Apple was, unsurprisingly, "to provide the high quality for viewing content without what people dislike about foldable displays: The distracting crease, the plastic feel, susceptibility to scratches and dents."

Apple started with what it knew and built something better

Talking to me after the event, Srouji and Apple CEO John Ternus probably leaned a little too hard into criticisms of other foldable displays. Ternus claimed, "I think what you've seen out there are all these just kind of cheap-feeling plastic surfaces that don't feel good to the touch."

Having used most of the major foldable phones out there, I don't think Ternus was being fair. After all, Apple has built its first foldable on the shoulders of giants like Samsung, who all but perfected the display design. Did Apple take it further? Absolutely.

In the keynote, Srouji said Apple's engineers relied on "their learnings from iPad and Mac" to develop the biggest folding display innovation: the nano-texture display. This reflection-killing surface is what's primarily responsible for hiding virtually all visual remnants of the crease.

I don't think the nano-texture covering (a polymer with a special anti-scratch coating, by the way) has anything to do with the possible cracks; instead, they probably relate to the multi-layered construction, and both the display and the hinges' ability to withstand thousands and thousands of folds (or, in the case of these cracked Duo screen customers, one bad unfold).

Based on the images Srouji showed off, the iPhone Duo display is composed of 10 layers. Included in them is a back panel of titanium (yes, like on the Galaxy Z Fold 8 Ultra) for straightness and stiffness when unfolded. There are also two ultra-thin flexible glass layers above and below the OLED panel.

Instead of these floating freely, Apple uses an elastic adhesive to keep the panels together while also letting them slide back and forth over each other so that, Srouji said, "the panel doesn't distort over time".

Behind the panel is the hinge, which Apple says has one hundred components, some of which are attached to carbon fiber support plates.

Watch this space

My point is that Apple was about as open as a tech company could be about how and why it made the iPhone Duo display, and we have enough information to make at least one reasonable assumption: one or more of the thin glass panels cracked for a few iPhone Duo owners. After all, the OLED panel is about as flexible as a thin piece of plastic. The titanium wouldn't crack; even the nanostructured polymer is unlikely to crack. My guess is that either a manufacturing imperfection or misuse resulted in one or both of the glass panels cracking, and possibly damaging the OLED sandwiched between them.

There are other hypotheses, like some loose component inside getting sandwiched and cracking the screen or even hinge misalignment. I'm sure Apple is examining every possibility.

The iPhone Duo is the first iPhone with a significant moving part. It's the first with two displays. And the first where one is a flexible display that's expected to survive tens of thousands of folds. The complexity and quality assurance for building such a device is surely enormous. Apple is lucky to have had others like Samsung, Motorola, and Google go through a decade of perfecting the technology before it delivered the iPhone Duo. Still, that doesn't diminish its accomplishment. Nor does it lessen the risks of shipping a folding iPhone to people who don't always know when to stop unfolding. Maybe someone over-torqued it; maybe they just got a bad unit.

In the end, such issues are not entirely unexpected, but they do not, for now at least, indicate a larger issue. I think Apple did its homework here, and I'm simply not buying 'crackgate.'

As Europe emerges from record-breaking summer temperatures, and as the US prepares for midterm elections in November, data centers find themselves in the firing line on both sides of the Atlantic.

In the UK, water companies say data centers are using 6.6 million liters of precious public supplies every day; in the US, some voters are blaming rises in their electricity bills on, you’ve guessed it, data centers . Bloomberg has described the data center “backlash” as a “defining midterms issue”.

And so, a narrative has been allowed to develop – that too many data centers are being built, and too quickly, and worst of all in areas where water is scarce. No wonder some communities believe their water needs are being ignored in the name of advancing IT infrastructure to support AI.

In some cases, there has been genuine cause for grievance – in an earlier era, water efficiency was not always a design constraint on data center construction.

But the public debate is now running on figures which range from outdated to plain wrong. Some of the loudest claims overstate the problem by orders of magnitude. Others, including several the industry itself repeats, understate it.

Getting the numbers right matters – and most of all to the communities that host these facilities – because bad numbers produce bad siting decisions, misdirected scrutiny, and misplaced trust.

The No.1 myth about water use

“Every ChatGPT conversation drinks a bottle of water” is the most famous quote in the public discourse about AI infrastructure and data centers. But the claim is not supported by anyone – not even its author. Traced to a single academic paper - Making AI Less “Thirsty” by Pengfei Li, Jianyi Yang, Mohammad A. Islam, and Shaolei Ren of UC Riverside – the figure was a calculated estimate built from public data, not an actual meter reading from a data center.

But a 2024 Washington Post analysis ensured the figure went viral. Since then, Ren has indicated that the GPT-4 estimate was likely far too high and that the real figure is closer to 15ml per prompt in total, of which 5ml is on-site cooling water in a data center. Then in August 2025, Google found that a median text prompt of its AI tool, Gemini, consumed 0.26ml of on-site water – or about five drops.

Credible estimates still span more than a thousandfold depending on model, location, season, and, above all, what you choose to count. But nobody is talking about bottles of water anymore – the honest answer to How much water does a query use? is a range, and the range is wide because disclosure is inconsistent.

Not all data centers are the same

Critics think all data centers are the same when it comes to water and cooling – but that’s simply not the case. Different cooling technologies use different amounts of water. Google disclosed in 2024, for example, that one of its Texas data centers consumed about 10,000 gallons in a year – roughly two months of a single household’s use – while its largest water consumer, in Iowa, used around a billion gallons. Same company, same year, but a difference of five orders of magnitude determined almost entirely by cooling architecture and climate.

There’s another myth that water used by data centers is lost forever. But again, no two facilities are the same. A data center which withdraws river water, runs it through heat exchangers and returns it to the river, at a slightly higher temperature, clearly has a very different impact from one which draws water from an aquifer and evaporates the same volume into the sky. Water withdrawal (and return) and water consumption are two different metrics. Again, bad numbers fail to tell the full story.

Golf is greedier when it comes to water

But are data centers really such water hogs? In most countries, their water use barely registers. In the US, for example, direct annual consumption by data centers accounts for 0.3% of total public supply. By comparison, golf courses use approximately thirty times more water than every American data center combined. And agriculture dwarfs both: the UN’s World Water Development Report puts farming at roughly 70% of global freshwater withdrawals.

National figures, though, obscure local impact. That’s why a single data center drawing millions of gallons of water a day, arriving suddenly in one small municipality, matters more locally than a thousand different water uses spread across a continent. It’s understandable – but important to put in context. Data centers attract community scrutiny not because they are large water users nationally but because they are concentrated and visible.

More AI doesn’t mean more water

The good news is that more AI doesn’t mean more water. The newest data centers have closed-loop designs, immersion cooling, and reclaimed-water agreements, precisely because the thermal loads of next-generation AI demand it.

The three largest hyperscalers – Microsoft, Google, and Amazon – have all committed to being “water positive” by 2030, replenishing more water than they consume. The EU, meanwhile, requires annual water reporting for data centers above 500kW. Regulation has a crucial role to play going forward.

The thirst for facts

On today’s evidence, AI data centers are a small national water user but with a real footprint. However, new engineering, better planning and regulation are achieving results. What is not yet solved is the accounting of water use. Standardized, facility-level and independently verifiable disclosure of withdrawals, discharge, and consumption would cost the industry relatively little and build trust with local communities.

Right now, the bad numbers around data centers and water use create misdirected scrutiny and misplaced trust. That needs to change. Honest accountability matters if public debate and policy are to keep pace with AI's growth.

We've reviewed, rated, and ranked the best web hosting services.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Modern digital security frameworks were originally built to correct a flawed historical assumption: the belief that if a user or system was inside the company network, they could be trusted implicitly. Over the past decade, the rapid rise of cloud storage, subscription-based business software, and remote working destroyed that network boundary.

In response, our security models shifted focus. We began verifying the identity of the person trying to log in, inspecting the hygiene of their computer, requiring multiple verification steps during login, and continuously checking that they were who they claimed to be.

This model works well when a human is logging in to perform a task. But with the rise of autonomous AI agents, the real security risk no centralized at the login screen. It starts after access has been granted.

An AI agent does not behave like a standard piece of software. It reads a human instruction, pulls context from company databases, utilizes multiple sets of tools, chooses its own digital tools, connects to external software communication links, and triggers complex automated workflows.

Verifying the initial login simply tells you that the agent is officially recognized by your system. It tells you absolutely nothing about whether the next action the agent plans to take is safe, ethical, or correct, focus has shifted to context.

The evolution of the machine identity problem

IT teams have already experienced a simpler, quieter version of this challenge with non-human identities. For years, companies have used automated service accounts, security tokens, and digital certificates to allow different software programs to talk to one another.

Most security leaders are all too familiar with the risks these credentials present: accounts are created for a temporary project, granted permanent access to database files, and then completely forgotten long after the project has ended. They become silent, unmonitored backdoors into the network.

However, AI agents complicate this familiar problem by introducing the element of independent judgment.

A traditional service account follows strict, pre-written code. By contrast, an AI agent interprets a high-level goal, analyses real-time context, selects its own tools, and actively decides how to proceed.

As a result, security policies can no longer stop at the front door. We must protect every internal data or compute asset the agent can reach and monitor every single decision it makes.

Matching autonomy with executive control

To manage this risk, security leaders must separate two distinct concepts: how much independence an AI agent has, and what level of file access it holds.

The combination of these two factors dictates the level of control required. For example, a basic agent that merely summarizes long documents only needs limited reading rights and clear activity logs. However, an agent designed to update sensitive, production databases, access code repos, write new types of code, or alter business workflows requires mandatory human approval steps, strict operational boundaries, and a clear path to reverse any mistakes.

We can categorize these digital workers into 3 distinct levels of independence, applying tighter security controls as their autonomy increases:

1. First, there are Conversational AI Agents (GenAI), productivity agents usually built and managed by employees & teams as part of their productivity (e.g. managed agents, self hosted chat bots). When driving security, we must enable proper RBAC authorization, associated with the employee and team, ensuring access to non-sensitive or production-related data, enabling logging, and validating the response to ensure proper integrity.

2.Second, there are Agents Acting with Semi/Full Human Guardrails (AI Agents), such as customer support tools that draft emails or prepare database updates for review. Here, security teams must validate the parameters of the tools being used and require a human to approve any actions involving writing, deleting, spending money, or sending information externally. All above controls apply

3.Finally, there are Fully Autonomous Agents (Agentic AI), which reset passwords or trigger complex workflows on their own. These require the strictest boundaries, from visibility (agent architecture, tool usage, and more) and focus on active set of controls. The controls include strict rate limits, continuous real-time policy checks, network access restrictions, run time security controls (Public/Private Cloud) continuous identity and minimal authorization verification, and finally complete, unalterable audit trails.

The industry is leaning toward autonomous agents seeking human approval upon abnormal behavior, can that be triggered in your agent?

The need for real-time inspection

To enforce these policies, security teams need a dedicated vantage point within the network. They must be able to inspect the human prompt, the agent's response, the digital tools being used, and the movement of data while the work is actually happening.

This requires an architecture that can discover all active agent activity across the business, govern how those agents use internal tools, and maintain complete visibility over data flows and automated workflows.

The true value of this approach lies in context. When the behavior of an AI agent is viewed alongside network traffic, user activity, and cloud database logs, security teams can investigate incidents with clear evidence. They can instantly see which agent ran, who in the organization owns it, what data was moved, and which automated tool was called.

For business leaders, this means cleaner, more consistent governance across human users, software applications, and digital agents alike.

How to begin: Start with a single workflow

Securing this new landscape does not require a complete overhaul of your entire security department overnight.

Start by picking a single AI workflow that is currently in use or close to being launched. Map out its digital identity, its level of independence, the files it can access, the tools it uses, the points where a human must approve an action, and how you would undo a mistake.

This simple mapping exercise will quickly highlight where your security gaps lie. If you cannot see the agent's actions in real-time, approve its decisions, block its access, or explain its behavior after the fact, your security controls are simply too far away.

From there, build a baseline on how your secure your coding agents, AI harness (or AI orchestrator) up to utilizing of tools/skills/MCP and moving forward to data assets access and utilization of identity that will allow you to set the correct foundations and guard rails.

The core principle of a "never trust, always verify" security model must follow the agent into its daily work. The most meaningful decision is no longer who is logging in, but whether their next action should be allowed to happen, under these exact conditions, right now.

We've featured the best endpoint protection software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

UK marketers are entering a decisive phase. Digital advertising investment continues to grow, retail media and connected TV are becoming more central to media plans, and artificial intelligence is moving from experimentation into everyday campaign operations. Yet, the basic commercial question has not changed. Can marketers prove that each pound spent is generating measurable value?

That question is becoming more urgent than ever. IAB UK has forecast digital ad spend to reach £45 billion this year, with mobile accounting for over two thirds of such spend and online retail media emerging as one of the fastest-growing channels. The same research points to measurement fragmentation as a major industry concern, while AI and automation are now reshaping how campaigns are planned, bought, and optimized.

For UK businesses, the opportunity is significant, but so is the risk. AI can accelerate decision-making, personalize creative, optimize bids, and expose new pockets of demand. But if the underlying measurement framework is weak, AI does not solve the problem; it simply scales it. A model trained on incomplete, inconsistent, or poorly governed data can produce confident recommendations that move budget in the wrong direction faster than a human team can intervene.

The mobile blueprint for better measurement

This is why the mobile measurement playbook matters. Mobile advertising was forced to confront privacy constraints, fraud, cross-platform identity, and signal loss earlier and more intensely than most other channels. The result was a more disciplined approach to attribution. Cleaner signals, stronger validation, privacy-safe methodologies, and a willingness to combine deterministic, probabilistic, and referrer-based approaches depending on the quality of the available data.

Those principles now need to be distilled across the whole UK media mix. Retail media may sit closer to the point of purchase, but it still needs transparent standards if brands are to compare outcomes fairly across networks. Connected TV, for example, offers premium, brand-safe environments.

However, advertisers continue to struggle with fragmented measurement and the link between exposure and sales. Social, search, display and commerce media all produce valuable signals, yet too often those signals remain trapped in channel-specific reporting structures.

The cost of measurement blind spots

The commercial consequence is underinvestment. When marketers cannot connect exposure to outcomes with confidence, they hold back from channels that may be profitable.

Research we undertook recently with EMARKETER found that nearly six-in-ten (58.6%) marketers believe their organizations are likely underinvesting in profitable channels because those channels require better measurement. These results are highly relevant to the UK, where media fragmentation, privacy regulation and economic pressure combine to make every budget decision more scrutinized.

In this environment, marketers need to treat measurement as infrastructure, not as a reporting layer added after campaigns have run. That means aligning definitions across teams, removing data silos, validating traffic quality, establishing privacy-compliant identity and enrichment processes, and measuring incrementality rather than relying solely on last-click or platform-reported attribution.

AI raises the stakes

AI makes this shift more urgent. As new AI agents and tools move across ad workflows, from creative production and audience selection to bid optimization and campaign analysis, they are squeezing advertisers already under pressure to prove ROI. Our research showed that almost half of organizations (49.7%) said they were under significant pressure to demonstrate clear ROI across all channels, while a further 29.9% reported extreme pressure.

That creates a new accountability challenge. Marketers must understand not only what an AI system recommends, but also whether the data feeding it is trustworthy. That trust gap is already showing up in how marketers use AI. Poor data quality, driven by fragmented, incomplete, or unreliable signals, was the top-cited barrier to extracting more value from AI in the marketing stack, cited by 22.3% of respondents. Explainability, or the difficulty of validating what AI is doing, came second at 17.8%.

In practical terms, AI-ready marketing data needs to be clean, deduplicated, consented, connected, and governed. If customer interactions across app, web, retail media, CTV, and CRM cannot be stitched together responsibly, AI will optimize against a partial view of the customer journey. That may improve a platform-level metric but likely weaken overall business performance.

What marketing leaders should do next

Firstly, marketers need to define measurement standards before scaling AI investment. Marketers should agree which outcomes matter, how they will be attributed, where incrementality testing is needed and which data sources are approved for optimization. Otherwise, AI risks becoming a more sophisticated way to chase inconsistent KPIs.

Next, brands should assess partners on transparency. Publishers, retail media networks, ad tech vendors, and agencies need to explain how they validate signals, reduce fraud, manage privacy constraints, and connect upper-funnel activity to commercial outcomes. In a market where regulatory change and economic uncertainty continue to shape planning cycles, vague measurement promises are no longer enough.

Thirdly, marketers should maintain human oversight where decisions carry material commercial risk. Automation can improve speed, but human judgement remains essential when interpreting anomalies, challenging assumptions and deciding whether a recommendation fits broader business strategy. The goal is not to slow AI down, but to ensure it is acting on the right evidence.

Finally, UK marketers should look to mobile as a model for resilience. Mobile measurement evolved in response to privacy limits, signal loss, multiple app stores, device fragmentation, and fraud. Those conditions now describe modern digital advertising. The channels may differ, but the underlying challenge is the same: measure accurately when the environment limits what can be seen.

The blueprint for accountable growth

The next phase of growth in UK digital advertising will not be won simply by adopting more AI tools or shifting more spend into emerging channels. It will be won by the businesses that can prove what works, understand why it works and act on that insight quickly. Mobile-grade measurement provides a practical blueprint: privacy-safe, fraud-aware, cross-channel and built for imperfect data.

The message is clear. AI may transform marketing execution, but measurement will determine whether that transformation creates significant value. With budgets under scrutiny and digital media becoming more complex, the brands that build stronger measurement foundations now will be best placed to turn automation into accountable growth.

We've featured the best AI chatbot for business.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

In recent weeks, we have seen eye-watering amounts of personal data exposed by criminal actors, in an attempt to either extort organizations for ransom payments or sell the information for financial gain.

In the UK, three major airports suffered data breaches that exposed information relating to nearly nine million people. In the fintech sector, Revolut mistakenly disclosed a wealth of customer data to scammers impersonating government officials.

While in North America, hackers reportedly offered scans of nearly 153 million US and Canadian driving licenses for sale on the dark web.

These incidents highlight the risks and vulnerabilities organizations face when they accumulate and retain large volumes of valuable personal data.

Beyond the obvious need for greater internet security and protection against AI related attacks or synthetic identities, it raises one other fundamental question: do companies really need to collect and retain so much personal data?

More data, more risk

Identity checks can typically involve companies collecting names, dates of birth, addresses, document scans and facial images. That data is then stored, and may be copied across different suppliers and systems in the hope that existing security protocols stay robust.

As a result, organizations may unintentionally be creating honeypots. If breached, this data could be used to facilitate a variety of illicit activities such as fraud, impersonation and social engineering. The more information an organization collects, the more places it is creating data exposure risks.

Data accumulation thus significantly increases a company’s security risk. As data’s value increases, breaches are becoming more lucrative for attackers and more catastrophic for businesses. As AI tools continue to become commoditized and easier to use, breaches may become even more frequent unless we find new ways to verify legitimate information without a person’s entire underlying dataset being at risk.

Verification doesn’t require lots of data

Many of the free online services we use everyday are at risk from disruption because they fundamentally ask for too much data. These services do not need to know 10+ pieces of personal information about an individual - yet, because this is still collected, the internet’s overall business model is vulnerable.

Authentication primitives that preserve privacy can offer a way forward, helping to protect the future of the internet. They enable people to access free content and services without revealing unnecessary personal details. For example, an age-gated service may only need proof that someone is above a minimum age requirement, or a government service may only need confirmation of residency.

All of these core aspects can be verified without needing to access and store someone’s full dataset. The only necessity is asking the right questions at the right time – in a way that ensures what needs to be verified is in fact legitimate.

Data minimization can therefore become a security control and risk mitigation tool. The less information a service collects and retains, the smaller the attack surface for illicit actors. By asking only for the information needed at a particular moment, organizations can attest what matters without creating an unnecessary, high-risk store of personal data.

Prove the fact: keep the details private

Modern zero-knowledge cryptography can help address this challenge. It enables individuals to optionally store digital credentials via information from physical, chip-enabled ID cards or government digital identities like mobile driver’s licenses through an app, without making it accessible to third parties.

An individual can then use their digital ID to prove things like age, nationality and unique document ownership without revealing their identity or giving that data to a third party. Zero-knowledge cryptography can verify these claims while keeping the original information private.

When proof generation takes place on the individual’s device, sensitive information can remain local while the receiving service gets the proof it needs. Holding less reusable identity data can limit what an attacker obtains if that service is breached.

Preserving privacy: why this matters

Beyond the security benefits outlined, reducing data retention will also become essential as AI makes it harder for us to distinguish what’s real, and what data to trust. At the same time, advances in computing power could mean that private data encrypted well today could be harvested and decrypted easily in a post-quantum future.

Zero-knowledge proofs are not the finished article, but this should form a key part of a broader shift to cryptographic systems that is going to be fundamental to how we verify real information it can trust, and prepares companies for threats that were barely conceivable a decade ago.

In a world where both personal data is becoming more valuable, harder to trust and secure, holding less will become one of the most essential security measures.

We've reviewed, rated and ranked the best Identity Theft Protection software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

One of AI’s biggest attractions is its ability to take time and effort out of everyday work. Tasks that would once have demanded hours of attention, such as turning meetings into notes or producing a first draft of a piece of content, can now be completed in minutes. These practical benefits help explain why AI has moved so quickly from experimentation into everyday business use.

Yet, as AI becomes more established in the workplace, a more complicated picture is emerging. AI is changing where thinking happens. When a system can produce an answer almost instantly, the challenge shifts from creating the output to deciding what to do with it.

Employees are no longer spending as much time getting to a first answer, but they are spending more time checking whether the output is accurate, and critically, good enough to inform a decision. This creates a different kind of cognitive burden, moving from production to verification.

AI-generated outputs can appear remarkably polished and persuasive, even when they are based on incomplete information. An answer can sound convincing while overlooking a crucial piece of context or relying on incomplete information. And as AI becomes increasingly better at sounding confident, mistakes may look plausible enough to escape scrutiny and feed into business decisions without being challenged.

When confidence outpaces accuracy

This is why discernment is becoming one of the most important skills in the AI-enabled workplace. The future workforce needs the ability to apply context, experience and judgment to an output that appears authoritative. Someone who understands a market or process deeply may immediately notice that something does not add up, while another person may accept the same answer because it is presented clearly and confidently.

However, the problem is that individual discernment does not scale well, and it is unrealistic to expect every employee to catch every questionable output, especially as AI begins to support more decisions across a business. Human review is indeed essential, but relying on individual vigilance is not a robust or sustainable operating model.

The challenge clearly magnifies with the growth of agentic AI. When AI moves from suggesting to doing, the consequences of a wrong answer become more serious because the error can travel quickly into a real business process. That makes robust governance increasingly important as agentic AI becomes more embedded in how businesses operate.

Better inputs lead to better decisions

Many businesses are investing heavily in making data available to AI, but availability is not the same as readiness. This is where many AI projects begin to struggle. Organizations too often focus on the model because it is the most visible part of the technology, while overlooking the quality of the information beneath it.

The reality is that most organizations will soon have access to similarly capable models. Consequently, what will increasingly differentiate them is not the AI itself, but the quality of the data, governance and operational discipline that sit behind it.

Making data work for AI means ensuring that information is current, trusted and structured in a way that allows AI to generate outputs that reflect how the business actually operates. Without those foundations, organizations spend more time validating AI than benefiting from it.

Poor data pushes additional work back onto employees, who need to spend more time checking outputs and resolving inconsistencies - weakening any productivity case. And the data backs this up. Whilst 74% of frontline employees are now regular AI users, 41% of those report increased cognitive load alongside productivity gains.

Making governance part of everyday operations

This is also why AI governance needs to move beyond policies and principles. If organizations want AI to deliver business value, they need to answer a series of practical questions, such as: Who owns an AI-assisted decision? And where should human judgment remain in the loop?

As AI adoption grows, those questions will increasingly bring the CFO into the conversation. Finance leaders will be asked not only to control spending but also to demonstrate that AI investment is producing profitable value. That will ultimately require greater visibility into where AI is being used, how decisions are being made and whether the costs of validation, oversight and correction are being properly accounted for.

It will also mean recognizing that governance is becoming a lever for cost efficiency. When AI systems are grounded in well-structured, trusted enterprise data, they can find the right information more efficiently, potentially reducing token consumption while improving the quality and consistency of outputs. In other words, strong governance can help lower the cost of running AI tools at scale and improve the return on AI investment.

The competitive edge is still human judgement

As AI becomes a standard part of how businesses operate, simply having access to the technology will no longer be enough to stand out. Instead, the greater advantage will come from how organizations are able to bring together reliable data, strong governance and human judgment to make better decisions.

AI can accelerate the route to an answer, but it cannot decide how much weight that answer should carry. That responsibility still belongs to people, supported by the right data, governance and decision frameworks.

We've featured the best AI website builder.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Polymath Michael Polanyi famously noted, “We can know more than we can tell,” a paradox that’s now showing up and being solved on the warehouse and retail floor.

Frontline workers possess deep, tacit knowledge of how operations run, mastering practical workarounds, managing exceptions and localized efficiencies, and situational problem-solving. This knowledge never makes it into an official manual, yet real-life experience and multiple academic studies say that up to 85% of company knowledge is tacit.

Because this knowledge relies on muscle memory, experience, ad hoc conversations during lunch and notes, traditional top-down standard operating procedures (SOPs) inherently fail to capture it. Tacit knowledge enables employee competence, autonomy and job satisfaction, but can be unevenly distributed or lead to gatekeeping to the detriment of others. And when experienced or long-serving employees retire or move to a new company, that tacit knowledge is lost.

For years, businesses assumed their SOPs were rock-solid, but the rollout of frontline AI tools, such as AI knowledge agents trained on SOPs and other workforce onboarding and training materials, is bringing long-overlooked knowledge gaps and outdated guidelines to the surface.

Agents are ushering in a quiet evolution in how enterprises capture institutional knowledge and build connected frontline onboarding and training programs. It comes as workplace culture, learning and development take on new urgency with organizations increasingly understanding their importance for workforce AI transformation.

The new knowledge economy

Instead of relying on centralized teams to write SOPs, the new connected frontline knowledge economy is going bottom-up by leveraging AI tools to crowdsource the “corporate brain,” with SOPs acting as the “nervous system” where knowledge is a two-way street.

By empowering the people stacking shelves, processing orders, talking to customers, and running the warehouses to flag inconsistencies, gaps and workarounds, companies are initiating a proactive audit and refresh of their workforce operations data quality.

Workers can actively signal knowledge gaps through their interactions with AI assistants, and those tools are routing the signals into dynamic, intelligently automated updates for training programs and SOPs. Workforce and operations leaders are finally able to turn undocumented, implicit frontline expertise into structured, actionable micro-learning, better procedures, and shift from one-off onboarding to ongoing everboarding.

When frontline contributions are recognized and shared, the workforce becomes better equipped to serve today’s highly informed customers, whether they are stepping into a shop or making an online purchase.

As David Autor, an MIT economist and author of a widely cited paper on the Polanyi paradox and employment said, “The tasks that have proved most vexing to automate are those demanding flexibility, judgment, and common sense—skills that we understand only tacitly...tasks that cannot be substituted by computerization are generally complemented by it.”

AI with all the senses

Organizations can and should move beyond text heavy documents and process guidance. Broadly speaking, learning and development departments already adopt a mix of text, images, videos, and group discussion. This approach reflects how we learn and should be the preferred method when it comes to employee onboarding, everboarding, and capturing implicit knowledge.

Modern AI agents are multimodal, meaning they can receive and output intelligence via audio, text, image, and video. A frontline worker should be able to verbally highlight a gap, suggest an update or snap a photo of a shelf with the AI agent comparing it to the visual merchandising planogram and provide step-by-step corrective instructions.

And with smaller models, on-device inference is possible, meaning no data leaves the device or company network. Latency is eliminated, an additional layer of security is added, and workers and customers get the intelligence they need in the moment.

Moving from knowing to doing

The development, implementation, and ongoing guidance agents require cannot be done alone, in the same way that text heavy, top-down SOPs are no longer sufficient. The top-down methods that brought organizations to this point will not sustain them for tomorrow.

According to a report by MIT, businesses that partner with AI providers have an AI success rate of 66% compared to those who prefer internal development alone (33%). The best way to succeed is to move forward with partners who have industry expertise, AI models, agents, and developer building blocks. They can support the design, configuration, onboarding, and management of AI systems in retail, warehouse, and logistics settings.

And AI transformation cannot be done with an “IT alone” mindset. Stanford’s Digital Lab finds that a top factor (43%) for AI success is executive sponsorship that enables a culture of experimentation and permission to fail. AI’s value comes from the algorithms, the technology required to implement, but mostly it comes from rethinking the people and culture component.

Polanyi’s “knowing versus telling” gap has been brought to light and is finally being closed on the frontline. By crowdsourcing the corporate brain, businesses are capturing tacit knowledge and transforming it into customer and worker advantages that make work better every day.

We've featured the best AI chatbot for business.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

As AI use expands globally, questions are emerging about where training data originates, who owns it, and how organizations keep sensitive information secure. This is the domain of sovereign AI: the protection of data, its uses, and its results, which is becoming increasingly important as AI moves from proof-of-concept to production deployments spanning entire countries and continents.

Beyond where data is stored, sovereign AI is really a question of who controls an organization's or region's AI capabilities. In practice, this requires control over the entire infrastructure stack, both software and hardware, along with policies governing who can view data, manage infrastructure, and install software.

Because data protection laws and practices vary widely across geographies – even within Europe – the governance of data flows and processing pipelines must be understood and followed.

While sovereign AI is often discussed on a country-by-country basis, the same rules and procedures apply to enterprises. For them, questions of data location, ownership, and transmission are even more complex.

Location is only part of the story

Much attention has focused on where a data center is physically located, but a closer look at what a data center actually comprises is needed when assessing sovereign infrastructure.

Physical location is usually easy to determine, but many providers run multi-cloud infrastructures in which data centers are linked by high-speed networks, and a cloud provider may shift workloads between its own facilities when capacity is tight. Organizations need assurance that their applications and data stay where originally intended.

If the proprietary data behind AI workloads sits with a cloud provider, the end-user may not know the exact location of the storage devices. Additionally, putting location aside, control is the critical concern: if one jurisdiction owns data stored in another, who decides what it can be used for and whether it can cross borders? The question grows more complex when a company has offices in multiple countries or regions.

Traditional workloads like ERP, inventory and HR systems typically keep data in a corporate database accessed remotely. AI is different: users are widely dispersed, while inference increasingly takes place on edge devices such as smartphones, connected vehicles and industrial systems. These devices often need access to the underlying data and trained model outputs, rather than receiving a stream of rendered content.

As models and proprietary data move between central infrastructure, regional data centers and edge environments, they may cross multiple networks and storage systems. Who owns that data in transit, especially when it crosses borders, is a question every organization scaling AI inference needs to answer.

Evaluating sovereignty layer-by-layer

AI today demands substantial compute, high-performance data input/output, and high-speed networking – components that must work together under a common set of laws, processes and procedures. An organization's AI infrastructure might combine powerful servers in one location, edge devices in another, and cross-border communication between them.

And then there's the data itself: where does it live, is it centralized, and in what format?

Ownership matters just as much as location. Data ownership might sit with an enterprise business unit, a compliance function, or the C-suite. Looking beyond the data, considerations include: who decides which data center runs a given training job, especially when a centralized scheduler manages many facilities for utilization and performance?

Can that automation be overridden? Should certain data be restricted to certain training uses, and does making specific data available to certain providers create enterprise value?

Because training data is drawn from sources around the world, enterprises need a clear governance framework for deciding which sources are trusted, who approves their use, and how regional regulatory or political requirements shape those decisions. That framework should establish accountability at both an organizational and individual level, with transparent records of where data originated, how it was used and whether the necessary permissions or licenses were secured.

The same principles should extend across the infrastructure stack. Organizations need defined ownership of the training environment, model weights and deployment decisions, alongside controls that determine who can administer these systems and from which jurisdiction. AI users may not need visibility into every technical detail, but they should understand the provenance, governance and location of the systems handling sensitive data and intellectual property.

Sovereign AI depends not only on where infrastructure is hosted, but on maintaining auditable control over the data, models, platforms, and people involved throughout the AI lifecycle.

There is also the challenge of demonstrating that an AI system’s outputs are grounded in legitimate training data, while ensuring that the data remains protected and accessible only to authorized parties across the infrastructure stack. Verda, a European AI cloud provider, offers one example of how this can be addressed in practice. Its infrastructure is hosted in Nordic data centers and allows customers to configure hardware and software around their operational and regulatory requirements.

The platform supports organizations across the private and public sectors, with controls designed to help address requirements including GDPR, the EU AI Act and sector-specific rules. These include confidential virtual machines, hardware-enforced trusted execution environments, and encryption for CPU and GPU memory, as well as data in transit and at rest.

The broader lesson that we can take from Verda’s success is that sovereign AI depends on embedding governance, security, and regulatory control throughout the system lifecycle, rather than treating data location as the only measure of sovereignty.

Operational sovereignty means more than compliance

AI is now critical infrastructure for enterprises worldwide, and compliance with local laws must be an architectural concern from the start, rather than being addressed after a data center goes live. From design through implementation and ongoing workloads, a sovereign data center must continually meet its operational sovereignty commitments.

Once online, sovereignty may help data centers remain operational through commercial disruption like recessions, geopolitical shifts or conflicts, and technical changes – new chips, models and architectures.

The concept of sovereignty is about resilience rather than eliminating every external dependency, with extreme events like war threatening to destroy infrastructure. It is about understanding where dependencies exist, deciding which ones are acceptable and avoiding unnecessary concentration of risk.

Technical decisions in design and operation can help avoid vendor lock-in, allowing a data center to keep running even if a business relationship ends. Relying on a single supplier for critical technology carries risk, since shifting geopolitics can bring tariffs or restrictions that block business with suppliers from certain countries.

Every data center operator depends on an ecosystem with a limited number of server, storage and power suppliers. Supply chains are global even when a facility sits in one geography, which optimizes cost but means not everything is controllable – so, understanding where risk concentrates matters. With many thousands of components from hundreds of suppliers, a smooth operation may still lean on a small set of vendors.

Open standards and formal standardization reduce this risk by making interoperability less dependent on any single supplier, and by allowing workloads to move between technologies and environments. In turn, this builds confidence in sovereign data centers as resilient infrastructure.

Building trust through control, not assumptions

Trust in sovereign AI depends on organizations knowing who can access their data, infrastructure and production environments, under what circumstances and with what oversight. Privileged access should be tightly controlled, clearly assigned and documented, with responsibility sitting at an appropriate level such as the CIO or Chief Compliance Officer.

This is particularly important when AI workloads involve proprietary data, prompts, model outputs or fine-tuning datasets that could reveal commercially sensitive information about an organization's products, customers and internal processes.

Policies and employee training remain important, but they should be reinforced through technical and operational controls. Organizations need clear assurances that their data will not be retained, reused or incorporated into provider model training without explicit permission.

Access controls, encryption, logging, workload isolation and data-retention settings can help enforce these expectations, while documented processes should show who controls each part of the AI environment, where supply-chain risks exist and which regional laws govern the data.

With this in mind, governance is becoming as important to AI infrastructure evaluation as performance, scalability and cost. Sovereign AI providers must be able to demonstrate, rather than simply claim, how they protect data and manage workloads across the full hardware and software stack.

For organizations increasingly reliant on AI as critical infrastructure, transparent ownership, auditable controls and consistent processes will be central to building confidence and supporting broader adoption.

We've featured the best AI tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Imagine a NASCAR team where leadership decides the winner is the car that burns the most fuel. The crew chief tunes every engine to run rich and wasteful. Drivers ride the gas and the brake at once. The telemetry shows record fuel consumption. The car finishes dead last.

For a stretch this year, wasteful consumption metrics drove AI measurement programs. Companies stood up internal leaderboards ranking employees by how many tokens they burned. A Meta employee built one, crowned the top "token legends," then quietly took it after the numbers leaked externally. By spring the trend had a name: tokenmaxxing.

By the start of summer the inevitable backlash arrived, with companies like Uber acknowledging they’d already spent their entire AI budget for the year in a matter of months.

More recently, the inevitable swing toward using the minimal number of AI tokens (tokenminning) leads to the same problem. If leadership defines winning as using the least amount of fuel, the winning car would never leave the starting line.

Burning or saving tokens should never be the end goal. The objective is shipping more products, at higher quality, with the same or fewer people. The job of a leader is to define the most strategic path toward the finish line, and based on that definition, build metrics that support it.

Why leaderboards miss the mark

A vanity metric is any number that looks great on a slide, leaderboard or dashboard but doesn't change how you run your business. In the AI space, even highly technical and complex metrics can become vanity metrics.

Here’s a couple I’ve seen lately, each a different way to look busy GPU utilization rates that brag about how much raw compute you’ve hoarded, regardless of whether it's solving real problems. AI features shipped per quarter, where leadership slaps a sparkle-emoji text summarizer on every legacy product to pad the release notes.

Metrics like these can feel rigorous, but they're detached from the end work itself. When leaders don’t relate metrics to end goals, engineers may race toward the nearest countable item.

The missing race map

Work completed only matters if it’s the right work. When leadership celebrates "95% adoption" of GitHub Copilot because the licenses are deployed, they can mistake superficial access for deep engagement. They assume productivity is skyrocketing. They're measuring who adopted the tool, not who is effectively using it to improve outcomes.

Developers might be generating more code faster, but if the tool isn't deeply integrated into a refined process, PR sizes inflate and the bottleneck simply shifts to human code review. Cycle time stays stagnant, or worse, bug rates go up because humans are rubber-stamping AI hallucinations.

Close the disconnect: tie depth of use directly to outcomes and business goals. If you have 95% "adoption" but your cycle time hasn't dropped by 30% or your rework hasn't decreased, you don't have high use. You just have an expensive new software distribution.

How to set the finish line

A single number rarely tells a story. If I can produce twice as much code but it takes twice as long to ship, I haven't gained efficiency. I've created more work. The right question isn't how much AI you're using — it's whether the work is actually getting better.

For engineering teams with more established workflows, start with the outcome you want to move, then use a framework like DORA or SPACE to check whether you got there. The framework is the instrument, not the goal. DORA tracks the health of your delivery pipeline — how often you deploy, how long changes take, how quickly you recover from failures, how often changes cause incidents.

SPACE tracks the full picture of developer productivity across five dimensions, including outcomes and satisfaction. Neither was designed for AI, which is exactly why they're useful: they’ll help you measure whether AI made the work better, not whether AI got used.

Pair one of those frameworks against your AI tool usage and your business outcomes like cycle time, rework rate, cost per change. A 25% increase in AI adoption that correlates with a drop in delivery stability isn't a win. The 2025 DORA report found that's exactly what many teams got when they deployed AI into underprepared systems. The tools amplified what was already there. Good engineering foundations got better, fragile ones got more fragile.

For roles outside engineering — marketing, finance, operations — build from first principles. Start with one outcome you're already trying to move: response time, pipeline conversion, forecast accuracy. Establish the finish line before AI enters the workflow. Measure toward that same outcome after. If it moved, figure out whether AI caused it and whether you can repeat it. If it didn't, you've learned something too. Picture a support team that wants to cut first-response time.

They baseline it at nine hours, then let agents draft replies with an AI assistant. Two months later response time is down to three hours, and reopen rates haven’t budged — so the speed isn’t costing them quality. That’s an outcome they can name, measure and repeat. It never needed a leaderboard.

Keeping the finish line in sight

The goal of a metric is to manage toward an outcome. If you're doing more work on the metric than on the outcome, you're upside down. You're looking at the gas tank, not the finish line. So before you stand up your next AI metric, ask one question: what outcome would move if this number did? If you can’t name it, you’re not measuring progress — you’re watching the fuel gauge. Define the finish line first; everything else is just telemetry.

We've featured the best AI website builder.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

I like the idea of AI taking care of boring tasks, including sorting through bills and financial paperwork. I can even see the appeal of it helping with shopping. But asking an AI assistant for financial advice and simply handing over the keys to your bank account are different, and this distinction is becoming a problem for some of America's biggest banks.

According to a new Axios report, financial institutions are encountering AI agents while they access customer accounts and seem to be doing more than requested by users. JPMorgan, Chase, Bank of America, and Morgan Stanley are all taking steps to protect their customers, but it's a sobering aspect of AI that raises real questions about how much to trust the tech in the world of finance.

While there's no confirmed epidemic of AI assistants emptying savings accounts and the reports don't establish any financial losses, it still makes me nervous to think that we've reached the point where banks are having to worry about whether an AI assistant has decided to improvise with somebody's money.

There's been a real push to encourage the use of ever-more independent AI agents. Every new demonstration seems to involve an assistant that can accomplish increasingly complicated tasks while its owner sits back and watches. That's impressive when the task is organizing a vacation itinerary, but less appealing when the assistant has access to the money that would pay for it.

AI experiments with money

It's notable that it isn't like AI agents are hacking their way into accounts, exactly. Customers might voluntarily share their login credentials with the AI to give the software access to financial information that banks have spent decades trying to protect. I've done it for my own tests, but it's always been simply a way for the AI to look at very specific elements of my finances, never to touch them.

Banking websites can be frustrating, financial paperwork is tedious, and managing money requires more attention than most of us would prefer to give it. An assistant that promises to handle the irritating parts sounds wonderful. But AI agents have a habit of interpreting instructions rather creatively. And while asking it to reduce your monthly expenses and getting a list of subscriptions you might want to cancel is great, giving that same assistant permission to make financial decisions means you're trusting its interpretation of a vague instruction with money you actually need.

That should make people uncomfortable. These systems can misunderstand requests and make incorrect judgments. We've all seen chatbots produce an answer that sounds perfectly reasonable until you discover they've invented half of it, and adding the ability to take real-world actions makes those mistakes considerably more consequential. The fact that banks are already encountering agents whose behavior raises questions about authorization suggests that those limits deserve much more attention.

Keeping AI outside the vault

I wouldn't go so far as to say banks should block AI assistants altogether. There are plenty of good reasons to let people connect these tools to their accounts, and I don't particularly like the idea of banks deciding which services customers are allowed to use.

But I'd be much less comfortable letting that same assistant actually do anything with my money I don't ask for. I'd want it to ask permission every single time, with no room for interpreting something I said last Tuesday as authorization to make a transfer today.

That might defeat some of the appeal of autonomous AI agents, but I'm perfectly fine with that. Saving a few seconds isn't worth the possibility of discovering that an AI assistant has made an expensive decision on my behalf. Most importantly, someone needs to be responsible when an agent makes a mistake, because telling customers to take it up with the chatbot isn't going to cut it.

The frustrating thing is that none of this should come as a surprise. AI companies are racing to give their agents more independence, often before they've worked out how to prevent them from making mistakes. We're supposed to be impressed that an assistant can complete a task without asking for help, but that becomes much less impressive when it does something the user never wanted.

CISOs and CIOs should not just chase the latest hype around AI that can escape sandboxes and morphing security threats, but bet on tried and true ROI of identity security.

It’s demonstrably true that the information security scene is transforming as agentic AI accelerates operational speed and potentially exposes new attack surfaces. Most AI providers have reported their latest generation models are so powerful that they are capable of escaping sandboxes and hacking organizations. Regardless if you believe this is hype, truth or genius marketing, what matters is this: information security leaders should rigorously check the basics of their discipline.

Yes, CISOs need to be ahead of the curve but this can’t be at the cost of dropping the ball on security and identity essentials. I say this because, all too often, even in sophisticated and otherwise mature enterprises, the latest breaches are still abusing the lack of essential controls, via credentials and permissions the agents should never have been able to reach.

And sometimes, thought leaders and experts attempting to understand and communicate rapid technology changes, are creating more confusion and distracting from first principles.

‘AI washing’ – the practice of overstating the importance of AI in a product, project or strategy – is real. We need to do more to demystify data security and separate hype from reality and place a higher value on specificity when we talk about risks.

The identity challenge

The industry's instinct to frame this as an "AI security" problem is understandable, but it is not actionable. For CISOs trying to reduce real risk, the more precise frame is identity.

For an example of what I’m talking about, consider onboarding. When people join an organization, they are granted privileges based on their needs, seniority and roles. A major part of contemporary information security is about staying on top of identity management, but that’s not easy.

In hypergrowth, companies hire very quickly. But IT leaders also need to stay abreast of many other people who need access to systems. These include freelancers, consultants, suppliers, auditors and work-experience temporary staff. That gap (unrevoked, over-permissioned, forgotten access) is exactly what autonomous AI agents exploited in the Hugging Face breach earlier this year.

The attack was novel. The underlying vulnerability was not. NHIs and human identities share the same failure mode: access granted, never reviewed, never revoked. That is the identity challenge, and it predates AI by decades.

A complex puzzle

IT today is complicated. CIOs stay in their roles for between three and five years. CISOs (and security staff generally) are in high demand, and organizations understandably turn to outsourcers for help. It’s easy for basic management disciplines like maintaining identity hygiene to get lost in the mix, such are the comings and goings of personnel.

And things are getting harder. Who owns an AI identity? Not every enterprise employs an identity specialist, so who is in charge when the identity is non-human? The CIO? What about the CISO? HR? How does the chief risk officer fit in?

It’s easy for tasks to fall between gaps. And even if onboarding is immaculate, who is running offboarding and the impacts on identity? The consequences of that question going unanswered are well documented. Verizon's 2026 Data Breach Investigations Report found that exploitation of vulnerabilities has overtaken stolen credentials as the top initial access vector for the first time in 19 years.

But credentials still drove the privilege escalation and lateral movement that followed. A vulnerability opens the door; an unmanaged identity walks through the building unchallenged.

And in the agentic AI era, who will be in charge of managing those NHIs that often have extraordinarily high permissions and license to roam? The Hugging Face breach answered that question in practice: nobody was. One over-privileged, unmonitored non-human identity provided cluster-admin access across two environments the moment it was compromised.

The AI was the intruder. Vulnerabilities have surged as an entry point precisely because AI agents are now extraordinary zero-day discovery engines, finding in hours what took human attackers weeks. But identity remains the common denominator: had proper least-privilege, just-in-time access and NHI governance been in place, the breach would have been contained even after the vulnerability was exploited. You cannot patch your way out of an identity problem.

Back to basics (and learning from the past)

If I think about a lot of the breaches I’ve seen, a lot of them relate to these seemingly obvious gaps that become chasms. It’s the left- behind orphan credentials, the ‘puppies with two owners’ and other grey areas that cause problems. ‘Hey. I thought you had this?’, people hear… when it’s too late.

To improve, we need to learn from the past, because we all know the famous quote (of uncertain authorship) about those who ignore history being condemned to repeat mistakes.

A relevant analogy here is cloud computing. Some of us lived through the datacenter wave, then the cloud wave hit. The Capital One breach of 2019 (over 100 million records exposed through a misconfigured AWS role) became the defining lesson of that era: cloud created vast new entitlement sprawl, and nobody was governing it. That breach, and others like it, drove the creation of Cloud Infrastructure Entitlement Management (CIEM) to bring least-privilege principles to cloud permissions at scale. AI is the current wave, and the pattern is similar.

Agents are being deployed with extraordinarily broad permissions, credentials are being embedded in pipelines without rotation or scoping, and ownership of who governs those entitlements is unclear. We will need the same focused response we built for the cloud. Not new acronyms, but the same rigorous application of least-privilege, just-in-time access, and entitlement governance, adapted to the new environment. And whatever wave follows AI will present the same challenge again. Identity is the constant.

Every wave reshapes the technology landscape, but the question of who (human or machine) has access to what, under what conditions, and for how long, never goes away. We just keep failing to answer it before the damage is done. If you weren't there, speak to a cloud veteran because so much of security today has to be about assembling teams of all talents. So use them and build consensus with clear allocation of who is responsible for what and a handover plan for when people leave or swap roles.

The current environment is replete with risks. Consider what AI has done to the act of building software: a non-coder like me can now spin up an app or deploy an agent in an afternoon. That is genuinely transformative, and dangerous. The population of people creating software that handles credentials, calls APIs, and interacts with production systems has exploded overnight, and most of them have never thought about identity hygiene in their lives.

Every time you use an LLM, you are storing credentials somewhere. As that population scales, so does the attack surface. We will see an avalanche of badly architected agents and apps built fast, deployed faster, and governed by nobody. It’s easy to offer up credentials to the world just as misconfigured AWS accounts led to major problems a couple of decades ago.

Today, everybody is excited about the abuses of agents running wild. These are hot stories, and yes, we should be concerned. But really, hackers are going to hack, just as they always have, and they will exploit any loopholes we offer up. The Verizon 2026 DBIR showed vulnerabilities are now the number one initial access vector, but identity is what converts a foothold into a catastrophe.

So many foundational security elements remain neglected like storing static credentials in plain sight. We can still do a lot more with Zero-Trust principles and enforcing least privilege to mitigate risks and reduce impacts. The tools exist and the discipline exists. CIEM was built precisely because the cloud created entitlement problems nobody had anticipated, and it worked, for organizations that adopted it seriously.

The same rigor applied to NHI governance and agentic AI today will determine whether the next wave of breach headlines features your organization or your competitor's. Let’s get the basics right and ensure we can walk at the same time as we try to run.

We've featured the best antivirus software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

AI has moved from pilot projects to the core of the finance function. Deloitte’s CFO Signals shows that close to 93% of CFOs now use AI across key operations, a dramatic shift from the tentative experimentation of just two years ago. The pace of adoption is undeniable but the organizations making the most meaningful progress won’t be the ones automating for the sake of it, they’re being deliberate in their choices.

Finance is full of tasks that look automatable but rely on context the data doesn’t capture. A spike in supplier costs might resemble a contract issue until you know it reflects a strategic shift agreed in a leadership meeting.

A variance might look like a warning sign until you understand it’s tied to a planned operational change. AI can highlight anomalies, but it cannot intuit the organizational nuance behind them. That gap between what the data shows and what the business knows is where trust can break down.

Finance teams need to know not only that AI can identify an issue, but when they can trust its interpretation and when human judgement needs to step in. This will be central to the next phase of AI adoption.

AI isn’t replacing judgement, it’s relocating it

I should be clear, the shift happening in finance isn’t about completely removing human judgement from the process, but slightly shifting where that judgement is used. Rather than having people spend their time checking every transaction or working through every variance, more of that initial analysis can be handled by AI, leaving finance teams to set the parameters, decide what needs attention and step in when something doesn’t quite fit.

That change is already reflected in how CFOs are using the technology. FTI Consulting’s Global CFO Survey found that nearly 70% of CFOs now rely on AI-driven insight alongside experience-based judgement rather than in place of it. So this is already a process that’s happening. In practice, judgement is still very much part of the role, but it is being applied further up the process, where the bigger decisions are made.

Trust shouldn’t be an afterthought

As we allow AI to take on more responsibility in finance, the question is no longer whether it can complete a task but whether finance teams are truly comfortable relying on the result.

That matters when dealing with sensitive data and decisions where a small error can have real financial or regulatory consequences. An AI tool that drafts a report is one thing. An agent that recommends a supplier, flags unusual spending or takes action on a payment is another.

Trust needs to be built into how these systems work, rather than added later through a policy document or training session. Finance teams need enough visibility into how an output was reached to challenge it when necessary, alongside clear boundaries around what an AI system can do independently and when it should hand a decision back to the team.

The model isn’t the bottleneck, the data environment is

The biggest obstacle to AI in finance is rarely the model. It’s the data underneath it. Finance teams have spent years stitching together information from accounting systems, expense platforms, procurement tools and spreadsheets. If those systems contain duplicate suppliers, inconsistent coding or outdated information, AI won’t fix the problem, it will simply accelerate the consequences.

Data quality needs to come before more ambitious AI projects. Clean, consistent and accessible financial data gives AI something reliable to work with. Fragmented data makes even sophisticated systems harder to trust. Much of the most important AI work happens before deployment: connecting systems, cleaning data and establishing a single, accurate view of financial information.

Once that foundation is in place, AI can do more than automate administration. It can help identify anomalies, understand spending patterns and give finance teams better information when decisions need to be made.

The real value sits in the unglamorous work: speeding up invoice processing, tightening expense categorization, spotting spend anomalies early or making planning data easier to surface. Starting with focused, practical use cases lets teams see how AI behaves in their own environment, where it genuinely helps, and what needs to change before expanding its role.

It’s a more realistic route than trying to transform the entire finance function at once. The organizations making progress are solving specific problems well, then scaling once they know the system works in the messy reality of their own data and workflows.

The finance team becomes more strategic

The real opportunity with AI is not to create a finance function that does less. It is to create one that spends less time on administration and more time on decisions that matter.

As AI takes on repetitive work, finance teams can focus more on analyzing the business, identifying risks and opportunities, and challenging assumptions. Instead of waiting for month-end to understand what happened, they can move towards more continuous analysis and decision-making.

AI does not remove the need for financial expertise. It makes that expertise more valuable, because people still need to decide what questions to ask, whether an answer makes sense and what action should follow.

The goal isn’t to automate everything

AI adoption in finance shouldn’t be measured by how many processes can be handed to an agent, but by whether teams can make better decisions with less administrative effort.

The key for CFO’s is knowing where AI can act independently, where it should make a recommendation and where human judgement is still needed. The finance teams that get this balance right will move beyond AI experimentation and turn it into genuine gains.

We've featured the best IT automation software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

UK organizations are already feeling AI's effect on the bottom line: a recent study found that 66% of UK enterprises are experiencing significant AI-driven productivity improvements, with leaders reporting that AI is freeing up time specifically for strategic decision-making, not just automating routine tasks.

That shift matters as, despite advances, analytics remained a retrospective or heavy human driven insight tool for long – reactive and slow to respond. With deterministic rule-based architecture, hard code, and static workflows. It worked well enough in predictable use-cases, but not in in dynamic, data-intensive scenarios.

This is changing rapidly, as AI re-imagines traditional analytics into decision intelligence, and autonomous AI goes even further, making decisions, scenario planning, and activating context aware actions with limited or no human instruction based on autonomy dial. Where traditional automation executed predefined workflows, agentic AI systems working in squads analyze data from varied sources to make a chain of complex decisions.

For instance, an AI underwriting agent typically estimates a prospective borrower’s creditworthiness, tweaks loan pricing, routes paperwork, alerts supervisors to compliance concerns. decisions once made by people, now made by systems working in squads.

The scale of the opportunity is significant. The decision intelligence market in Europe, estimated to be $4.35 billion in 2025, is steadily advancing towards a projected value of $20.61 billion by 2035. UK organizations could be well placed to capture a meaningful share of that value if they move deliberately.

Drawing the line between AI and human judgement

The central challenge for any organization adopting decision intelligence is not the technology itself, it's deciding who gets to decide what and when. Several organizations lack a formal mechanism for deciding the extent of autonomy they should allow their systems. This frequently leads to ad-hoc practices, without any limit’s enforcement or governance of autonomy.

The fix is to re-invent or re-imagine clear business decision flow rights into the system from the outset - what an agent may act on independently, and what it must escalate to a person. Going back to underwriting agent example, it can reasonably price a loan or route paperwork on its own, but if it flags a compliance concern or encounters an applicant whose profile falls outside its constraints data, that decision should stop with a human, not get resolved by the agent guessing its way through an edge case.

The future enterprise is not about replacing human judgment but augmenting it through intelligent delegation. Enterprise Decision Intelligence systems, fueled by domain aware foundational models and machine-consumable context, can autonomously handle high-volume, full explainability, data-intensive decisions across supply chains, customer operations, pricing, and activation ecosystems.

Human expertise, meanwhile, remains indispensable for governance, ethical oversight, exception handling, and decisions where business context outweighs algorithmic certainty.

Profound potential, profound risks

The potential benefits are substantial. For instance, when beauty and fragrance retailers have implemented an AI decision engine that determined the right time and proposition for engaging each customer, it earned a double-digit ROI through strategic AI collaboration.

However, AI-based decision intelligence systems come with concerns such as cybersecurity risk, compliance failure, lack of transparency and accountability, and occasionally, improper outcomes.

Even the entry cost, especially when agentic AI is involved, can be quite high and can be optimized through enterprise intelligence foundation models. Above all, full implementation needs to be iterative, involving not only a technology upgrade, but also operating model changes, deep integrations, specialized talent, and a robust ethical and governance framework to build transparency, accountability, and trust in these systems.

As organizations build experience and confidence, they will be able to progressively transition from reactive analytics to proactive decision intelligence. The ones that get the human-AI division of labor right first, not just the ones that move fastest, will be the ones that capture the advantage.

We've featured the best AI tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Quantum computing still feels like uncharted territory, yet the threat it poses is significant. On the one hand, it promises advancements in areas such as financial modelling. On the other, it has the potential to undermine the encryption standards that protect governments, businesses, and critical infrastructure around the world.

People often draw parallels with AI, but the comparison is imperfect. AI is transforming how organizations operate. Quantum computing could fundamentally challenge the cryptographic foundations that protect sensitive data and digital communications. While the timeline remains uncertain, organizations need to prepare now. The earlier they start to assess their cryptographic exposure and planning for post-quantum security, the better prepared they will be when the threat materializes.

Quantum is speeding up

Investment in quantum computing is growing, with major economies such as the US already directing significant funding in the technology. The UK has also previously committed up to £2 billion in quantum innovation and aims to become the first country in the world to manufacture and deploy quantum computers at scale.

This is driving the development of quantum computing, which could bring significant benefits. By tackling complex calculations that would take traditional computers years to complete, the technology could optimize energy and transport networks or streamline global supply chains. Yet, like AI, it also poses a significant cybersecurity risk.

Most of the today’s public-key cryptography, including RSA and elliptic curve cryptography (ECC), relies on mathematical problems that would take conventional computers a significant amount of time to solve. A sufficiently powerful quantum computer could solve these at an exponentially faster rate, potentially undermining the encryption used to protect sensitive information.

While there is no definitive date for when quantum computing will be able to break widely used encryption – often referred to as ‘Q-Day’ – the cybersecurity threat posed by the technology is already here.

The threat is already here

Threat actors are deploying a strategy known as ‘harvest now, decrypt later’, collecting and storing encrypted information today with the intention of decrypting it once quantum computers reach sufficient capability. This means information does not need to be vulnerable today to be at risk.

This is particularly concerning for sensitive, long-lived data that needs to remain confidential, including government and critical infrastructure data, intellectual property and confidential business information. Data intercepted now could remain secure, only to become exposed once quantum technology advances later.

Ultimately, organizations – regardless of industry – cannot afford to wait until Q-Day arrives. Now is the time to transition to quantum-safe security. Yet, it can be difficult to know where to start.

Protection starts now

The first step for organizations is to understand where they are currently exposed. In practice, that means identifying the information that needs long-term protection and understanding the cryptography used to secure it. A cryptographic audit can help bring these dependencies to light, enabling teams to decide which systems and data need attention first.

Those findings can guide the transition towards quantum-resistant security, in which post-quantum cryptography (PQC) will play a central role. The National Institute of Standards and Technology (NIST)’s Post-Quantum Cryptography standards provide a clear starting point for organizations planning to migrate their existing environments.

Other approaches have more specific applications. Quantum key distribution (QKD) uses quantum mechanics to share encryption keys in a way that can reveal interception attempts. It requires specialized infrastructure, making it more suited to high value uses than to widespread deployment. The right approach will depend on what an organization needs to protect, how long that information must remain confidential and the systems it relies on.

AI has demonstrated just how quickly emerging technologies can transform cybersecurity, and quantum computing could be next. Organizations cannot afford to wait for the threat to fully materialize, and should begin the transition today by mapping exposure, testing alternatives and establishing migration priorities. The sooner they do so, the more time they will have to manage the transition before their information becomes at risk.

We've featured the best endpoint protection software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

As the AI buildout continues and models improve from generation to generation, fears persist over how useful this technology will actually be in practice and whether AI will reach a ceiling before it makes a meaningful impact.

Although roles and responsibilities, and jobs, are being replaced in the name of AI, does that mean it's changing things for the better?

Energy scarcity

Microsoft CEO Satya Nadella was speaking at the World Economic Forum in Davos in 2026 about the impact of AI on the global economy and the business world.

In his remarks, Nadella acknowledged the tensions at the heart of the AI expansionism that has captured not only big tech but the vast majority of the corporate world.

For AI to continue growing at the pace it has been – not just in terms of the compute required to train the models but also for inference as usage grows – we'll need plenty of data centers to churn through tokens. And for these tokens to be created, the AI pioneers will need plenty of energy.

Encountering bottlenecks

Burning as much electricity as AI companies plan on doing so for AI has caused alarm among the public – particularly as much of this has demanded infrastructure built close to populated areas in towns and cities.

Nadella acknowledged there has been a productivity gap between the hype around AI and the meaningful impact on people's lives – warning that without something to point to, unrest will bubble up and boil over into something more obstructive.

As time passed, we're seeing more tensions rise as people – and even legislators – object to new data center projects due to concerns over the impact on the energy grid, the environmental damage, and local communities.

Mini-LED vs OLED has been a battle in the TV world for a few years now. While mini-LED always provided more brightness and better bright room viewing and OLED delivered better black levels, thanks to its self-dimming pixels, that has changed in recent times, with both TV techs managing to cover more bases.

Both LG and TCL make some of the best TVs you can get, with the former delivering some of the best OLED TVs and the latter offering up some of the best mini-LED TVs. Both also deliver great value for the price you pay, especially TCL.

The LG B6 is one of the best OLEDs I’ve tested this year, offering great performance for a more affordable price. Now, the TCL C7L (this is the UK / AU model — the TCL QM7L is the US equivalent) has arrived in our testing lab so I thought I’d compare the C7L, the brand’s mid-range SQD mini-LED for 2026, against LG’s most entry-level OLED, as they’re available for similar prices.

Can the mini-LED hold its own against OLED? A quick note that we’re using the LG B6E, a cheaper B6 that performs the same other than missing some AI picture modes, and with a different back panel.

Brightness

Flagship and mid-range OLED TVs have seen a brightness boost over the years, with one example being the near 1,000-nit peak HDR brightness increase between the LG G4 (1,489 nits) from 2024, and its 2026 successor the LG G6 (2,475 nits). It’s taken until this year for entry-level OLED to get a brightness boost: and it’s not as significant as that.

The LG B6E measured 808 nits HDR peak brightness (10% window pattern) in Filmmaker Mode in my tests: a nice boost over the B5’s 668 nits. However, this 808 nits pales in comparison to the 2,083 nits peak HDR brightness achieved by the C7L in its Filmmaker Mode. Mini-LED has always been the brighter tech, but do these results reflect real-life?

In practically every scene I tested, the answer was a resounding yes. With both TVs set to Dolby Vision Filmmaker, in a scene where Lawrence and his guide are by a well in the desert in Lawrence of Arabia, the white sands and blue skies were much more impactful on the C7L, showing off the brightness advantage. The B6E did a good job with the picture, but next to the C7L it looked dim in comparison. Switching to the B6E’s Cinema Home mode did give it a boost in brightness, but still the C7L’s picture dazzled in comparison.

This was also apparent in HDR highlights in other movies. In Speed Racer, the brilliant white of the Mach 5 was much punchier on the C7L than the B6E, though the gap in luminance wasn’t quite as obvious as it had been in Lawrence of Arabia.

For bright-room viewing, neither TV was perfect, with both showing mirror-like reflections. The C7L did a better job in brighter conditions, thanks to that added brightness — it's the clear winner for pure brightness, and for visibility in strongly lit rooms

Contrast and dark scenes

OLED’s advantage when it comes to darker, more contrast-y scenes is its self-dimming pixels. As it doesn’t rely on a backlight and it can turn off individual pixels, this produces much deeper black tones that most mini-LEDs can achieve, and with precise borders between light and dark. However, more mini-LED TVs are demonstrating better backlight control and more impressive black tones.

In my time testing the C7L, I’ve been impressed with its black levels: especially for a mid-range mini-LED. Yes, there’s still some blooming going on, but it’s greatly reduced from the previous generations of TCL’s TVs I’ve tested before — not that long ago, this would've been premium performance. Still, when it comes to black handling, OLED is king.

Watching any space scenes from Project Hail Mary, the B6E clearly produces richer, deeper black tones that are consistently strong throughout. The C7L, while rendering strong black tones of its own, don’t quite have the same consistency, as some shots of the stars in space demonstrate blooming, a common occurrence with backlit TVs.

When it comes to contrast and shadows however, there’s a bit of give-and-take. Watching The Batman, with its array of contrast-rich scenes, the B6E clearly produces a more balanced image thanks to its deeper dark tones and well-handled bright tones. The TCL’s picture however is more impactful thanks to those punchier highlights. As Bruce works at his desk in the Batcave, the dark crevices and wood of the desk blend well with the lamps and monitors on both TVs.

In fact, some black tones can actually look crushed on the B6E, leading to a loss of shadow detail, such as the wall at the rear of a subway platform in The Batman. The same surface is much more legible on the C7L. Switching from Filmmaker to Cinema Home on the B6E does rectify this, without compromising significantly on dark tones.

The Batman has proven to be a challenge for backlit TVs in my time testing, due to its low brightness. While The C7L did a great job overall, occasional blooming would appear behind bright objects, which was more noticeable when viewed in dimmed conditions. For dark scenes and dark rooms, the B6E proved to be the better choice.

Color

Color is probably the most contentious subject when it comes to mini-LED vs OLED, as both tech can deliver great colors in their own way, and I found this to be true putting these two TVs side-by-side.

Watching The Super Mario Galaxy Movie, both TVs deliver vibrant, bold colors that accurately delivered the pop you’d want from this movie. The C7L’s colors were punchier overall, thanks to that higher brightness, but the B6E’s colors looked richer, with more depth. This was obvious whenever the Lumas were on screen, where they had more of a 3D-look on the B6E, but were much brighter on the C7L. Mario and Luigi’s overalls showed that both TVs delivered great colors, but the B6E’s looked that bit more 3D and solid.

Switching to Speed Racer, both TVs again captured the cartoon-ish, candy-colored aesthetic of the movie well. The C7L had the more shimmering colors, but the B6E’s were more true-to-life, as shown by the grass in Trixie’s flashback early in the movie. I also found that different colors looked better on different TVs, with reds and greens standing out more on the B6E, but yellow and orange looked better on the C7L.

One advantage that the C7L did have was in its color temperature accuracy. The B6E demonstrated a green tint in comparison, which can be an issue for some OLEDs. In isolated viewing this isn't so obvious, but next to the C7L, it was noticeable. In the subway scene in The Batman, the surroundings had a green hue to them on the B6E visible above, but the C7L’s cooler color temperature helped render more accurate gray tones.

This scene was the most obvious case of the B6E’s green tint I found during my testing, and it wasn’t legible in other movies such as Project Hail Mary and Speed Racer.

In terms of which TV delivers the better color, they both have their strengths and both have great color reproduction overall, but if I had to pick one, it would be the B6E purely because I prefer the richer color profile. But the C7L’s colors were very good.

The best buy

When it comes to value for money, both these TVs tick plenty of boxes. Both TVs have a feature set up there with the best gaming TVs, with 4K 120Hz, variable refresh rate, auto low latency mode and Dolby Vision gaming supported, although the B6E has four HDMI 2.1 ports to the C7L’s two — but they both have razor-sharp performance.

Picture quality comes down to preference, and both deliver adequate built-in sound, though both could benefit from one of the best soundbars.

When it comes to price, it's close in most places. For the 55-inch models I tested, you’ll be looking to pay $799 for both in the US, with the QM7L going for $100 cheaper in the larger 65-inch size. In the UK and Australia, the C7L is better value, with the 55-inch costing £899 / AU$1,495 while the 55-inch B6E costs roughly £1,099 in the UK (though it has dropped to £999 before) and the standard B6 costs £1,199 / AU$2,495.

Honestly, you can’t really go wrong with either TV. The TCL is the much better pick for bright rooms, while the B6E is the better choice for darker, cinematic viewing and I really am torn on which is the better pick, as this was such a close contest.

Purely from a price standpoint, the TCL is the winner (especially when you go beyond 65 inches — mini-LED's price advantage grows as sizes do), and is the more practical choice for a bright room.

But if you can get the B6E for a good price and you'll watch in a room with controlled lighting, then the better contrast and color depth make it my pick.

Thinking of buying a new TV?

Try our TV size and model finder! You tell it how far you sit from your TV, we'll tell you what size to buy based on viewing angle advice from image quality experts, and we'll recommend our three top TVs at that size for different prices.

The late Apple co-founder Steve Jobs had a reputation for being ahead of the curve. At the time, Microsoft was heavily pushing touchscreen computing in an attempt to innovate and embrace new possibilities when interfacing with machines; Jobs dug his heels in and refused to take Apple down that same path.

Swimming against the tide

Steve Jobs was speaking during the 'Back to the Mac' media event at Apple's corporate headquarters while introducing Mac OS X 10.7 Lion.

Instead of adding a touchscreen to the MacBook, which Jobs saw as something of a gimmick, Apple focused on multi-touch gestures using the trackpad and mice instead of adding a touchscreen into the fabric of the display. The reason, according to Jobs, was simple: using this technology would quickly lead to strain, fatigue, and the loss of precision the longer a user was navigating using touch.

He put his foot down despite immense industry pressure at the time in the form of the emergence of new kinds of devices such as the HP Touchsmart family.

Times are changing

With Windows continuing to introduce new touchscreen devices and 2-in-1s to the market, Apple stood firm on its resolve over the next few years – but the manufacturer experimented with bridge technologies like the touchbar, which was eventually phased out.

There has since been renewed pressure on Apple to embrace touchscreen and embed this in its devices – with the company now expected to break its stance for the first time and introduce a new touchscreen MacBook Pro with an OLED panel, according to rumors.

By nearly all accounts, we’re inching closer to a passwordless society. The FIDO Alliance, which develops new passwordless authentication standards and whose tagline is quite literally “reducing reliance on passwords”, now estimates that more than 5 billion passkeys are in use worldwide. In fact, more than three-quarters of global consumers have already enabled a passkey on at least one account.

The technology is a boon for the financial industry because it finally removes the sector’s entrenched dependency on knowledge-based credentials – reusable “secrets” that criminals can steal, guess, or extract through social engineering. Suddenly, being able to quote a user’s mother’s maiden name, the town they grew up in, or their several-digit password, isn’t enough – and that makes user access instantly more secure.

Passkeys also give customers a much easier ride when it comes to accessing their accounts, speeding up the process by allowing them to use a key biometric like a fingerprint rather than hopping through two or three screens and entering their “memorable information” each time they log on.

So far so good. However, according to research by Javelin, account takeovers still impacted some 6 million US consumers in 2025 – an increase of 18% on the previous year – with losses exceeding $15 billion. Countless other stats and analyses also support this trend, including The Federal Reserve, which reported a 36% increase in account takeovers between 2023-4 – and those are just the incidents that were brought to its attention.

A surface read of these figures suggests that passkeys are failing outright, but dive a little deeper and it’s clear that the problem isn’t with the technology itself, but how it’s implemented and safeguarded. Securing the front door doesn’t work if a fraudster can manipulate the locksmith into handing them a new key.

While a successful passkey authentication establishes control of a device-linked credential, it’s still on financial institutions to prove how that credential came to be enrolled and whether the session it grants access to remains trustworthy. Did the account holder genuinely intend to add that payee, recover that account, or move a huge sum of money? Passkeys are useful, but they were never designed to answer all of these questions on their own.

In finance, scale doesn’t always equate to risk

Back in August, researchers at Palo Alto Networks’ Unit 42 published details of three proof-of-concept attacks against passkey-protected accounts. They demonstrated how malware on one compromised device could authenticate itself without the account holder even being present. It could even interfere with the registration of a new passkey, or even expose the passkeys synchronized to that user’s environment. Seemingly, another blow for passkeys.

This is an alarming piece of research for sure, but best taken with a grain of salt. Delve deeper into the report, and it’s clear that every one of these scenarios began with a “poisoned” device, where malware was already present. The most serious cases also depended on a very particular combination of browsers and operating systems with preset security hardware.

Reassuringly, none of the cases had an “Enigma moment” where they broke the cryptography behind passkeys, nor did they uncover a “master vault” of credentials that could be ransacked in a single raid. That means attacks like these, where customers are pursued individually, would be far harder to scale than the kind of breaches we frequently read about in the headlines.

But “hard to scale” isn’t really much of a defense in finance. A criminal doesn’t need to steal access to thousands of accounts if one “whale” justifies the effort, and if successful, that single breach can spell doom for a bank’s reputation or regulatory standing – so warnings, while taken in context, should nevertheless be heeded.

Account recovery was, and remains, the weakest link

One criticism users often have of passkeys is, “what happens when I lose or switch my device?” It’s an inconvenience for customers, and it's a real headache for banks. Of course, banks want to make the experience of switching devices as quick and painless as possible, but issuing a new passkey isn’t like resetting a password. It will only be as trustworthy as the processes used to issue it.

If passkey recovery falls back on an email, a text, or a PIN sitting in a notebook somewhere, all an attacker has to do is clear that relatively weak hurdle. How can a finance enterprise know if the person claiming to have a new device isn’t just an attacker who has come across this information by chance or while targeting a specific individual? It relegates passkeys to basic, knowledge-based security again, undermining the very purpose they were created for.

A bank simply cannot recover confidence in a customer’s identity if it never established that confidence in the first place. During enrollment, when the bank verifies who an individual is and binds their identity to a trusted record, it’s usually done through a server-held biometric reference – something that is preserved even when the original device is lost.

If a situation arises where a customer needs to recover an account, the bank should always return to that same source of truth rather than create new “evidence” by asking for it through channels that may have already been compromised. So while recovery may not be needed months or years after onboarding, the efficacy of it is really decided on day one.

Authentication shouldn’t be confused with authorization

The line between authentication and authorization may seem invisible at times, but there’s a canyon between the two – particularly when it comes to moving money. Authentication establishes access, but authorization is what determines less tangible things like permission and intent. A passkey might prove the presence of a credential on the device, while a biometric or PIN check “releases” it and opens it for the user.

That confirms that the authenticator approved the login, but it doesn’t mean that everything that follows on that session is legitimate. Proving a user logged in legitimately at 10:19 doesn’t also prove that they meant to transfer $5,000 to a new payee at 10:26 – and that’s the grand canyon between authentication and authorization.

How the finance industry crosses that canyon depends on their ability to establish user intent in the moment. Of course, they can’t simply add gateways at every single point in the user’s journey – that kind of friction would be enough of an inconvenience for customers to start considering switching to a competitor.

But they can engage in step-up authentication and do what they can to establish contextual baselines. If a user logs on at 9:50 and checks their balance, like they have done countless times before at roughly the same time every few days, the passkey alone is enough. If they want to move money, add a new payee, or – perish the thought – recover their account, they need to reestablish trust by going back to the server stored source of truth outlined above.

Each request should warrant its own process for establishing truth depending on its level of sensitivity and what is at stake.

Putting passkeys in their place

When it comes to modern banking, passkeys absolutely deserve their seat at the head of the table. They strip away the traditional dependencies on knowledge that can be easily stolen, and replace it with device-based credentials that are far harder to crack from a social engineering perspective. On paper, passkeys should banish account takeovers, but in practice, there’s still work to be done.

Finance enterprises still need to know who was originally bound to a device-based credential, and they need to be able to separate access from intent. Passkeys can carry some of that burden, but not all of it. Increasingly, finance is on the front lines of the new war on trust, and they must do what they can to establish it, nurture it, and preserve it.

We've featured the best antivirus software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Ask most technology leaders anecdotally how their customer relationships are doing, and the answer sounds solid. Renewal rates hold up, and people are continuing to invest in tech to stay competitive. However, there’s a danger in getting too comfortable.

Recent research among UK IT decision-makers found that while the large majority rate their vendor relationships positively overall, almost half say slow response times are a persistent frustration, and 40% flag integration issues without a proper fix.

Leaders who only track the headline number miss the friction clearly sitting beneath it. This matters for two reasons: how customers find technology providers has changed, and so has what they need from them.

AI is reshaping discovery

Many consumers now use AI when making a purchase. A simple search of ‘best budget sofas’ will generate a list of pros and cons from Google’s Gemini. Whether it's a reseller weighing up a new supplier or an enterprise buyer choosing between platforms, the same is largely true for tech buyers. Somewhere between deciding they have a problem and picking up the phone, they stop and ask an AI tool what it thinks.

Forrester's State of Business Buying report recently found that AI tools are the single most meaningful source of information buyers use, ahead of vendor websites, sales reps and product experts combined. Gartner's research backs this up, finding that 45% of B2B buyers used generative AI during a recent purchase, mainly to gather information on vendors and products before any human contact. AI is the first step.

People still want a conversation before they sign anything, but if your reputation isn't holding up in the places AI tools are reading, you don't get considered for the conversation at all.

What tech firms can sometimes forget is that AI discovery matters to both current and prospective customers. It only takes a quick search from a so-called ‘happy’ customer to see a competitor's name where yours should be and start wondering whether their loyalty is misplaced.

To show up in the answers AI tools give, tech companies need quality owned channels like website copy, blogs and case studies, alongside third-party endorsements from reviews, press coverage, forum threads and comparison sites. AI often gives as much credit to what other people have said about a business as to what it says about itself. Marketing efforts need to be multi-channel and shaped around the specific questions a buyer is asking.

With GEO (generative engine optimization) still so new, the rules are shifting fast. Tech providers must know what’s being said about their brand in the places that matter, and keep closing the gap between that and what buyers need to hear.

Complexity is the real retention risk

Picture a UCaaS provider that's worked with a mid-sized accountancy firm since the pandemic, keeping partners and clients connected through voice, video and messaging as the firm went hybrid.

The provider does discovery well: strong owned content, credible reviews, regular mentions in the trade press, the kind of presence that shows up when an AI tool is asked to compare providers. Over the years, it's added feature after feature to the platform, each one an improvement on paper. Then, with no real warning, the firm gives notice.

The reason is rarely one dramatic failure. The firm grew, added a second office, and needed its comms platform to integrate with the existing practice management software. Instead, every new feature meant another login, another dashboard, another thing for a firm with no in-house IT team to configure.

Nobody at the provider connected this to billable hours, client responsiveness, compliance – outcomes the firm cares about. The technology hadn't got worse, but the firm was overwhelmed and felt its provider didn’t understand the business behind the invoice.

The accountancy firm's problem was never about willingness. It had no in-house IT team, and each new feature landed on the same handful of people already juggling billing, compliance and client work, which is typical for SMEs.

The Government’s 2026 skills assessment found that the skills needed for digital roles are changing 66% faster than in most other jobs, while smaller businesses have disproportionately less access to the training needed to keep pace. Firms are finding it hard to keep up, which is exactly why the burden of translating new technology falls to whoever supplies it.

Customers value providers who act as strategic advisors, simplify integration and focus on real business outcomes. In practice, that means:

• Treating integration as an ongoing part of the relationship, not a box ticked during onboarding.
• Translating new capability into the numbers a customer already tracks.
• Flagging complexity early, while there's still time to plan around it.

The winning supplier is the one who takes work off a stretched customer's plate. Get that right, and skills shortages become the reason a customer needs you more. Get it wrong, and the accountancy firm's story repeats: a customer who looked comfortable, right up until the moment they weren't.

Never get too comfortable

The suppliers who hold onto customers over the next few years won't be the ones with the most features. They'll be the ones who show up accurately wherever a buyer is looking, and who make the day-to-day easier for a customer that has less capacity than they used to.

That means auditing what AI tools already say about the brand, treating integration as ongoing work and checking in on the customers who look most settled, since they're often the ones closest to leaving.

We've featured the best customer experience (CX) tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

On 12 June, thousands of organizations discovered an uncomfortable truth about the AI tools they had started building into their businesses. They didn’t have control over it.

The US government issued an export-control directive affecting two Anthropic models. Anthropic responded by disabling them for every customer. The restrictions were later lifted, but that is hardly reassuring. Access disappeared and then reappeared because of decisions customers neither made nor controlled.

It’s a wake-up call that many claimed that they had after the Cloudflare outage… or the CrowdStrike outage… or Meta’s… or AWS’… or so on… And yet we were met with predictable surprise amongst an industry that has been so afraid of having potentially uncomfortable conversations around digital sovereignty.

We often talk about data-center locations, regional availability and compliance certificates. We discuss whether the server sits in York or New York, then congratulate ourselves for making a sovereign choice.

But if a foreign government can remove your service, an outage can take down your entire business, a vendor can change your operating model or a proprietary API makes migration economically impossible, the location of the server is almost beside the point.

Data residency is a comfort blanket

Putting a server in Paris does not magically make your architecture French. The server is only one part of the system and it may not even be the most important part. Sovereignty is about control over the complete operational chain. That includes the infrastructure, of course, but also the tools used to provision it, access it, deploy software to it and keep it running.

A business can satisfy its data-residency requirements while remaining completely dependent on foreign technology. A compliance box is checked and the organization believes it has reduced its risk because the data has moved, but the levers governing that data remain in somebody else’s hands.

The control plane matters. The developer tooling matters. The source-code platform, identity provider, orchestrator and AI assistant all matter. If any one of them becomes an immovable dependency, the organization is less sovereign than it thinks.

As an industry, we have focused on the physical location of servers because it is easy to understand. Technical autonomy is much harder, and it requires organizations to examine dependencies they have spent years pretending are somebody else’s problem.

Enterprises are stuck in a legal sandwich

If technical lock-in wasn’t bad enough, the legal situation makes it worse. Modern enterprises increasingly find themselves trapped in a legal sandwich, caught between rules that apply simultaneously but do not always point in the same direction.

Weave Intelligence demonstrates the example of the friction between the US CLOUD Act and the EU’s GDPR.

The CLOUD Act can require US-based technology companies to disclose data under their control, even when that data is stored outside the United States. GDPR, meanwhile, places strict conditions on transfers of European personal data to third countries and access by foreign authorities.

This creates an obvious problem for a European company using a US-owned technology provider. The data might be sitting in Frankfurt, but the company operating the service can still be subject to US law. If it receives a valid demand from the US authorities, the provider may face obligations in one jurisdiction that create serious legal and regulatory risks in another.

This is not just a US-EU issue of course, but rather an example of where residency is often second to complex legal definitions of data ownership. And whilst enterprises cannot solve this international law through architecture, they can stop making the problem worse by building every critical workflow around one provider, one jurisdiction and one set of proprietary services.

The legal sandwich becomes much harder to swallow when there is no realistic way to change what is inside it.

If you cannot leave, you are not sovereign

Digital sovereignty means retaining the practical ability to change provider, jurisdiction or deployment model without bringing the business to a halt.

Managed services allow teams to build quickly without worrying about every part of the underlying infrastructure. But each provider-specific service creates another dependency, and each dependency makes the exit door slightly smaller. This is the genius, and the true cost of the hyperscaler.

The test is very straightforward. If your provider disappeared tomorrow, could your teams continue building and operating software?

If the answer exists only in a 40-page exit-strategy document that nobody has tested, the answer is no. A genuine exit strategy must be executable and the business must be able to shift workloads, change providers or bring critical capabilities on premises without redesigning how every development team works.

Own the layer between developers and infrastructure

To achieve genuine technical autonomy, organizations must decouple software delivery from the infrastructure underneath it. This is the strategic role of the Internal Developer Platform.

An IDP provides a consistent layer between development teams and the different environments where their software runs. Developers follow the same golden paths to build, test and deploy, whether the destination is a local cloud provider, a global hyperscaler or an on-premise environment.

The infrastructure can change without forcing every developer to change how they work. More importantly, the organization retains control over the workflows that keep software moving. Deployment logic, policies, configurations and operational knowledge are not welded to one infrastructure vendor.

This turns portability from a paragraph in a risk document into a capability that can be exercised and tested. Organizations can move workloads between environments during normal operations, identify the dependencies that remain and address them before a crisis arrives.

Stop claiming sovereignty. Prove it.

Real sovereignty has to be designed, tested and maintained. Organizations should regularly move workloads between environments, test whether critical tools can run independently and expose the dependencies hiding behind their architecture diagrams.

Because the first time you test your sovereignty should not be during an outage, a legal dispute or a geopolitical crisis.

We've featured the best database software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The growing use of enterprise AI is reshaping decision-making, accountability and executive roles, forcing CEOs to rethink how their organizations operate from the top down. The result: the C-suite, traditionally the upholder of continuity and stability, is being redesigned.

Confidence in AI is increasing, but unlocking its full value requires organizations to overcome barriers around governance, skills and workforce readiness. By establishing clear guardrails, investing in talent and embracing new operating models, leaders can accelerate adoption while building trust across the organization.

Looking beyond resilience

Chief executives are no strangers to navigating persistent economic, geopolitical and regulatory pressures. However, the most progressive leaders are looking beyond mere resilience, they are starting to harness the power of AI to fundamentally reshape their leadership, business model and the organization that supports it.

A transformation is underway that is rewriting how global executives approach organizational design. Forward-thinking, "AI-first" leaders – those already realizing the performance gains of integrated technology, are no longer attaching AI to legacy structures. Instead, they are leveraging these tools to fundamentally redesign executive decision-making, decentralize accountability and pioneer entirely new paths for value creation.

AI is a shared responsibility

Consequently, AI is no longer a niche, back-office function. It has expanded to become a top priority for CEOs around the globe, triggering a strategic reconfiguration of the C-suite and forcing leaders to constructively reshape corporate identities and strategies. The challenge now facing these leaders is that progress here sits with organizational and behavioral change, not simply technology. This change is harder to achieve and softer to measure in the short term.

To accelerate operations, executives are shifting accountability deeper into the organization. This redistribution of power is coupled with a call for broader technological expertise; CEOs increasingly agree that all business leaders must become technology literate in their own domains.

AI competency is no longer optional. To navigate this rapid shift, organizations are elevating the role of the Chief AI Officer (CAIO) to a business-critical requirement.

Trust vs caution

Executive confidence in the strategic potential of AI is significantly rising, with leadership becoming increasingly comfortable making business decisions based on AI-generated insights. This trust is particularly pronounced among those spearheading technological integration who anticipate a substantial shift toward automation, expecting AI tools to eventually guide a major portion of daily operational decisions.

However, this executive optimism stands in stark contrast to the reality of daily operations. A distinct gap remains in widespread workforce adoption, as only a minority of employees routinely use these tools in their day-to-day roles.

This lag is not driven by a lack of confidence in staff capabilities – leadership largely believes teams possess the necessary skills. Instead, it points to deeper organizational hurdles, which involves a large cultural shift in the way organizations work and how employees behave.

This measured implementation is primarily driven by critical concerns surrounding governance and risk management. For modern business leaders, AI sovereignty, the capability to control exactly where enterprise data is stored, processed and accessed, is non-negotiable. This highlights the absolute necessity of establishing robust internal controls and strict operational guardrails before expanding deployment.

Success starts with people

The most successful technology strategies are human-centric. Modern enterprise leaders widely agree that long-term success hinges far more on widespread employee adoption than on the sophistication of the technology itself. Tools alone cannot serve as a silver bullet for business transformation.

Importantly, employees are not resisting change due to a lack of access to advanced tools. Adoption is stalled because daily workflows, performance incentives and organizational trust have not evolved at the same pace. Unlocking value at scale requires behavioral change, redesigned operational processes and a collaborative culture.

This shift demands a transformation of talent strategy. Over the coming years, a significant portion of the workforce will require comprehensive retraining to work alongside advanced systems in their current roles, while others must be completely reskilled to transition into entirely new functions.

Managing this shift requires tight integration between talent management and technology leadership, ensuring that human capital and technology strategy are developed in tandem.

Writing a new playbook

The AI era requires a new playbook. Organizations that have redesigned their core workflows and operating models around AI are already reporting greater revenue growth than their peers. For leaders aiming to replicate this success, here are five strategic moves:

1. Redesign the C-Suite – Restructure leadership for agility and clarity, formalizing roles (including the CAIO) and establishing AI competency as a baseline expectation for all functional leaders.
2. Embed AI across workflows – Move beyond isolated pilots to optimize end-to-end efficiency and effectiveness with embedded AI agents.
3. Customize the AI mix – Prioritize long-term differentiation over short-term ROI by developing a hybrid strategy that combines custom, foundation and specialized AI models.
4. Orchestrate intelligence – Focus heavily on the human element by aligning daily workflows, building new skills and actively fostering a workplace culture of trust.
5. Expect the unpredictable – View an "AI-first" operating model not as a final destination, but as a foundation for future disruption. The C-suite must remain agile and prepared for what comes next, whether it is quantum computing or another unforeseen technological leap.

The transition to an AI-powered enterprise is an organizational challenge that extends far beyond the IT department. It involves a corporate disruption that requires a new kind of leadership – one that is agile and focused on integrating people and technology to unlock growth and new value.

We've featured the best AI website for business.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Real-time has become something of a benchmark in IT. Teams see a problem, diagnose it quickly and, ideally, remediate it before the employee has time to raise a ticket.

Clearly, that's better than waiting for an employee to report a problem to the service desk. But real-time isn't necessarily preventative - if an issue occurs and we respond in milliseconds, we have still reacted to an issue.

And if the same thing happens again tomorrow? An automated remediation may mean the employee barely notices and an engineer never gets involved, which is valuable. But we're still fixing something we've fixed before.

Recurring problems carry a cost even when IT becomes very efficient at dealing with them. The cost shows up on both sides: employees lose time when technology doesn't work as expected, while IT spends time investigating and remediating familiar conditions. This is where Site Reliability Engineering (SRE) offers a way to break that cycle.

Bringing SRE to the endpoint

SRE came out of Google's approach to running large-scale production systems, applying software engineering principles to operational problems. At its simplest, it means defining the reliability a service needs, measuring it and reducing the repetitive work involved in keeping it there.

A Service Level Objective (SLO) puts a measurable target around that reliability, while error budgets recognize that no service will be perfect all of the time. These are established ideas in software and infrastructure, but they translate increasingly well to the digital employee experience.

The device, applications and services an employee relies on are now part of how the organization operates, and their reliability has a direct bearing on whether people can get work done.

A recent conversation with a partner put the challenge neatly. They said: “Tell me about emerging risk whilst it’s still at 0–2% and give me the evidence to prioritize. At 5–10% beyond baseline, the service desk will have already lit up.”

They were talking about deviation from their normal baseline, not the percentage of employees affected. They wanted enough evidence to spot deterioration early and decide whether it warranted attention before the consequences showed up in support demand. A millisecond reaction is good, but prevention is always better.

That depends on being able to see problems before they become obvious. Digital employee experience data can expose issues that conventional operational measures miss, giving IT evidence of what is happening before the service desk becomes the signal.

At a large Fortune 100 insurance organization, for example, the team discovered that approximately 25% of call-center downtime was effectively invisible to IT. By combining endpoint telemetry with proactive automation, the organization reduced ticket volume by 35% in a year and avoided more than 45,000 tickets through a single self-service automation.

Learn from what keeps coming back

Anyone who has worked in IT operations knows how much time can disappear into familiar problems. A large hospital network found that a small number of recurring issues were generating a disproportionate amount of service desk activity. They included problems with file synchronization, security services, CPU utilization, network services and collaboration tools.

By using endpoint data to identify and address those recurring causes, the organization reduced annual ticket volume by roughly 50,000 incidents and estimated savings of approximately $1.6 million.

This is where the SRE idea of toil maps particularly well to workplace IT. Toil is the repetitive operational work involved in keeping a service running, and automation can take a lot of that work away from engineers. But if the same remediation keeps being required, that pattern is worth paying attention to.

Sometimes automation will remain the sensible answer. Not every repeated condition needs an engineering project behind it. In other cases, the evidence may point towards an application, configuration, update, or another underlying problem that can be tackled more permanently.

Global investment management firm PIMCO provides another example of what changes when teams have better evidence to work with. Its infrastructure team used endpoint evidence and change correlation to reduce root-cause investigations that had previously taken days to minutes.

That speed matters, but so does what the team can do with the evidence: understand what changed, isolate the cause and get closer to eliminating a recurring issue rather than repeatedly treating its symptoms.

Knowing what happened after the fix

We have plenty of evidence that tells us an IT action happened. A script completed, an update installed, a device restarted or a workflow reached its final step. What matters next is whether it worked. Did the issue disappear? Did the affected experience return to its expected range? Was the improvement still there a day or a week later?

Subsequent telemetry can provide that answer. Without it, IT knows an intervention took place but has less evidence of the outcome or whether the improvement lasted. SLOs help by giving teams an expected level of reliability against which to assess recovery. A completed action is one data point, while the state of the service afterward is another.

AI makes this feedback increasingly important because more diagnosis and remediation will happen without an engineer examining every signal or action individually. If AI identifies a problem and initiates a response, IT still needs evidence of the outcome.

The workflow from insight to outcome matters here. As more of it becomes autonomous, validation and accountability need to remain part of it. Otherwise, IT could automate more activity without necessarily making the environment more reliable.

None of this argues against real-time observability, automation or an efficient service desk. We need them. How quickly IT reacts still matters, as does how much manual work automation removes.

For CIOs, the bigger gain comes when those capabilities lead to fewer problems requiring attention in the first place. That means spotting deterioration earlier, learning from recurring issues and checking whether interventions have actually held. We've become very good at fixing problems quickly, and now SRE gives us a chance to spend less time fixing the ones we've seen before.

We've featured the best laptop for programming.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

On September 14th, 2026, AI frontier lab OpenAI called on UK lawmakers to impose regulations to tackle the security threats posed by the rapid development of artificial intelligence. This follows an unsolicited attack from an OpenAI agent that broke the firewalls of startup Hugging Face earlier this year.

Cybersecurity isn’t a new issue for business leaders, but its significance is increasing in the age of AI. Clients trust them with information, employees don’t want to disclose their salaries or investors don’t want to make their investments public.

Without clear governance and processes on the use of AI, these are all at risk of being leaked. Doing nothing poses an existential risk for firms.

A new class of security risks

AI presents great opportunities to speed up repetitive tasks, like data reporting, leaving more time for employees to produce more strategic, quality work for clients. For an agent to become useful, it needs as much context and information access as possible, but this is where there is the greatest potential for something to go wrong if not managed correctly.

For instance, agents may have access to bookkeeping, reporting, document management and CRM systems, meaning a successful attack could expose highly sensitive client information.

The main issue is LLMs have difficulty distinguishing data from instructions. An agent can interpret instructions from an email, a document or a webpage, even if those are malicious. These are called prompt injection attacks. Basically hackers manipulate generative AI systems by feeding them with malicious inputs disguised as legitimate user prompts.

Another threat is data leaking. An agent can visit a website or perform a web search and potentially release confidential information, for example by incorporating company data into a URL or a search query. This could expose sensitive data to the Internet without the company or the agent noticing.

The risk becomes particularly acute when three things come together: an agent receives untrusted inputs, accesses sensitive information and takes action in external systems – the so-called Lethal Trifecta. A malicious instruction hidden in an email, document or webpage can potentially trigger actions at a later point using the agent's legitimate permissions.

Therefore, firms need to set strict guardrails around each agent to limit what data it can access, what capabilities it has and what actions it is allowed to take. This is essential to avoid information leaks or an AI agent going rogue.

The threat sitting inside a company

AI adoption is scaling across all levels of a business, from finance to marketing, sales to product development. New research from Turbotic shows 51% of business leaders fear employees are feeding sensitive company data into unauthorized AI tools and 36% are concerned about employees using AI without oversight. This is called Shadow AI.

When an employee prompts a public AI platform, it leaks sensitive client information and receives output that looks confident but could be inaccurate. This could lead to inconsistent tone, misaligned advice or data breaches. The problem isn’t the technology itself, but it’s the lack of traceability and human review. Placing too much confidence in AI tools without the right guardrails is dangerous.

Most of the time, employees don’t even realize the security issues that pertain to AI usage.

They need to understand that experimentation outside any approved workflow or data safeguard is a security risk. Sharing sensitive information with a public model or using content generated by AI without reviewing it risks greater exposure to compliance breaches, AI slop and reputational damage.

Most importantly, employees become responsible for output they cannot explain, verify or trace.

To remove this risk, leaders need to democratize AI by giving employees access and clear guidance on the tools they’re allowed to use, how to provide the right context to AI agents, set up processes to review AI output and establish any guardrails to prevent data breaches. This approach that we call Organizational AI will make a firm more capable from security and capacity perspectives.

Managing the risks of AI

Despite increasing warnings about AI security threats, fewer than 1% of organizations have fully operationalized responsible AI, according to a report from the World Economic Forum and Accenture.

Given the risks associated with free AI usage, having a responsible AI policy is now obligatory.

Without restrictions on basic access controls, employees may be able to use AI to surface information they shouldn’t have access to, from colleagues’ salaries or confidential investor data. Once an AI agent is connected to an organization's data, weak permissions are a liability and it can more easily surface accidental access.

Anything that relates to company data, client communications, or that influences professional advice needs to be controlled by clear governance. It’s the responsibility of leaders to provide employees with auditable and transparent boundaries, and step one involves providing clear guidelines on what, when and how AI tools can be used. Then comes explainability combined with governance and oversight to generate trust.

The critical need for Organizational AI

Firms need to get the fundamentals right: What data can AI access? Where does the input come from? What systems can it connect to? What actions is it allowed to take? When must a human review or approve the output? All of these should be clearly communicated to employees and baked into processes.

Organizational AI is implementing a system that runs across the whole firm that the organization can grow into for a better client experience, team development and security management.

Companies should start now. The existential risk of doing nothing is bigger than the risk of doing it badly.

We've featured the best AI website builder.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Facial recognition technology (FRT) is fast becoming part of everyday life in the UK.

In recent months, much of the media attention around FRT has focused on the police, as forces up and down the country roll out both retrospective and live facial recognition (LFR).

As deployment increases, so too does scrutiny. Concerns around the accuracy, privacy and efficacy, alongside debates over proportionality and whether widespread use is creating a “surveillance society”, all deserve serious consideration.

But while so much of the conversation around FRT focuses on police use, we risk overlooking the growing use of facial recognition by private organizations.

Public attitudes change depending on who is behind the camera

Public attitudes towards facial recognition are rarely determined by the technology alone. Where it is used, why it is deployed, and, crucially, who is operating it can all influence public levels of support.

There is an important distinction between FRT being used by the police to identify a wanted suspect – often for violent or sexual offences – and it being deployed by a retailer to identify someone suspected of previously shoplifting. Questions around accuracy, privacy, and the handling of biometric data are valid in both cases, but the purpose and context are very different.

This disparity is reflected in public attitudes. Earlier this year, Face Int commissioned an independent survey of 2,000 UK adults which found that almost three quarters (73%) support police use of FRT, compared with 54% who support its use in retail environments.

This does not mean that police use should evade scrutiny, nor does it necessarily mean that people oppose private sector deployment. What it does show is that public acceptance of facial recognition cannot be separated from the circumstances in which it is used, or from the organization operating it.

When police use LFR, its presence is typically made clear through marked vehicles and signage, often accompanied by advance announcements about when and where it will operate, and how data will be processed.

In retail spaces, its use may not always be as obvious, nor will the public necessarily understand what happens to their data. This is particularly sensitive: 57% of respondents to Face Int’s survey said they were concerned about how images of their faces are stored.

Moving beyond accuracy

Facial recognition is often discussed in terms of its accuracy, with many reports of individuals being falsely identified. Ensuring that the technology is accurate, robust and appropriately tested is essential.

But accuracy alone will not secure public trust. How facial data is collected, processed and stored, and governance surrounding its use, matter just as much. Indeed, 43% of the UK public stated that clear laws governing FRT would increase their trust in the technology.

In the UK, FRT is currently regulated through a patchwork of legislation, including the Data Protection Act and the Human Rights Act. This means that those deploying the technology must interpret several laws to ensure they are compliant, leading to inconsistent standards across the sector. In contrast, the EU AI Act specifically addresses biometric identification, with additional requirements to protect the public and their data due to come into effect in December 2027.

Earlier this year, the Government completed a consultation on a new framework for law enforcement use of FRT, but notably this did not extend to private sector deployments.

More recently, the Independent Commissioner’s Office (ICO) published findings from its work examining police use of FRT, stating that it was working with forces and police oversight bodies to give the public confidence in its governance and oversight.

What responsible use should look like

Greater scrutiny does not necessarily mean more restrictions. It should provide clearer expectations for how organizations capture, process and store biometric data.

Organizations must be transparent about when and why FRT is being used. Information about the purpose of a deployment, what happens to people’s data and who is responsible for it, should be readily available before the data is captured.

Privacy should also be built into FRT systems from the outset, rather than added after sensitive data has already been collected. Techniques such as hashing and other forms of biometric template protection can remove the need to capture facial images. This privacy-by-design approach should be a minimum expectation for those investing in FRT.

Additionally, organizations should only capture and retain data they genuinely need. Data relating to individuals who do not match a legitimate watchlist should not be retained, while clear retention period should apply to information that does need to be stored.

Robust security is also essential. Access should be restricted, with clear records of why, when and by whom data has been accessed. Anyone handling biometric data should receive appropriate training, and organizations must have processes in place for responding quickly and transparently to breaches.

Finally, technology should support human decision making, not replace it. While the technology is improving rapidly, facial recognition matches must still be treated as an alert requiring human review.

An opportunity for the new Government

The recent change in government and dissolution of the Department for Science, Innovation and Technology (DSIT) makes this debate particularly timely.

With DSIT’s responsibilities being spread across various departments and embedded across the Government’s wider economic and social agenda, there is an opportunity to develop a more coherent approach to facial recognition. After all, the implications of facial recognition extend far beyond the technology itself, spanning policing, the high street, data protection, security and economic growth.

The Government must ensure its response to the proliferation of FRT is cohesive, with departments working alongside regulators and the private sector.

The ICO’s examination of police use of FRT is a useful model for how scrutiny of the private sector could work. This could identify good practice and areas of concern, and inform clearer guidance or regulation.

Responsible businesses should welcome that scrutiny. Clearer standards give businesses greater certainty about what good deployment looks like, and give the public greater confidence that their biometric data will be handled appropriately.

The debate over facial recognition is not a binary choice between embracing the technology or restricting it. The question is how we can realize its benefits while ensuring its use is transparent, proportionate, accurate, and secure.

Police deployments will rightly remain part of the conversation, but as facial recognition becomes commonplace across the private sector, scrutiny must follow. Public trust will depend not only on what the technology can do, but on whether people have confidence in how it is used and how their data is protected.

We've featured the best authenticator app.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

We understand what it means to own a house, a car, a share in a company, a patent or a trademark. In each case, there is an established legal framework that defines the asset, establishes who holds it and enables it to be sold, transferred, financed or protected.

Data is more complicated. A business may have spent decades building a valuable body of data that it protects and depends upon, yet establishing precisely what constitutes that asset, and what rights are attached to it, is far less straightforward.

That uncertainty has commercial consequences. It can make data harder to value, license or share with confidence, leaving transactions heavily dependent on contracts and trust between the parties rather than an independently verifiable understanding of what the data is and how it may be used.

This is a challenge we have been exploring in the Isle of Man through the development of the Data Asset Foundations framework, which establishes the statutory basis for defined, governed data to be recognized as an asset.

With the framework now moving into implementation, a much broader question remains relevant for businesses: if we are serious about treating data as an asset, what does that actually require?

What makes data an asset?

The first point is that a data asset cannot simply mean any dataset.

For a body of data to function as a distinct asset, four fundamental questions need to be answered. What exactly is it? Not simply "our customer records", but a precise boundary: which information, covering which period, from which sources and, if the data has been worked on rather than simply collected, how it was produced.

Who may use it? What is the holder actually permitted to do, and what limits came attached to the data when it was gathered? Those limits do not disappear on registration; they travel with the asset as registration does not displace existing contractual, intellectual property or data-protection rights; those obligations continue to apply.

Who is responsible for it? There needs to be a named party accountable for keeping the asset in good order and maintaining its governance. How can others have confidence in it? A means needs to be established by which a third party can see that the asset has been independently assessed and understand its assurance status, without needing direct access to the underlying data itself.

These are relatively simple questions, but many organizations would struggle to answer all four consistently across the data they hold. Doing so requires businesses to move beyond broad descriptions of data and identify distinct bodies of information with defined boundaries, rights, responsibilities and potential uses.

That distinction has practical commercial value. Once there is greater certainty around what the asset is and the rights attached to it, there is a stronger basis from which it can be valued, licensed, shared or potentially used in financing.

Why the rules matter as much as the data

Perhaps the most important part of this shift is recognizing that a data asset is more than its underlying information. Its governance matters too. Typically, businesses hold data in one place while information about how that data should be governed sits elsewhere.

Its provenance, permitted uses, quality requirements and restrictions may be spread across policies, contracts, spreadsheets and other internal records. The relationship between the data and those rules therefore depends heavily on the organization maintaining it.

The Isle of Man's Data Asset Foundations framework provides the statutory basis for bringing the two together. The Data Asset Register, which is now being implemented as part of the framework's supporting operational arrangements, will record information defining the asset and its governance, including its identity, provenance, rights, permissions, responsibilities and means of verification.

Crucially, the underlying dataset itself will not sit on the Register. Instead, the information recorded will provide a consistent way of establishing what the asset is and the governance and rights attached to it.

There is a useful comparison with land registration. A land register does not determine whether a property is valuable or well managed. Instead, it provides certainty about what the property consists of, where its boundaries lie and who owns it, thereby reducing friction when conducting transactions based on the property.

The same principle has commercial relevance for data. Greater certainty will not create value where none exists, but it can make existing value easier to recognize and use, whether through licensing, sharing, financing or the development of new data-driven products.

For a potential investor, lender or licensee, that could make a material difference. Today, assessing a dataset may mean piecing together its provenance, permissions and governance from contracts, policies and internal systems. A defined and governed data asset can provide a clearer evidential starting point for that assessment, reducing some of the friction involved without determining the asset's value or guaranteeing a particular commercial outcome.

Making data an asset in more than name

For years, businesses have described data as one of their most valuable assets. But there is a difference between recognizing that data has value and creating the conditions that allow organizations to act on that value. Much of that challenge comes down to certainty: knowing what the asset is, where it came from, what rights attach to it and how those claims can be verified.

The Isle of Man's Data Asset Foundations framework is intended to provide that greater clarity. By creating a clear bedrock of uniform understanding, organizations' self-described most valuable asset can move from an internal resource to a defined, registered and governable asset, providing a much stronger basis for valuation, financing and commercial use.

As AI increases both the demand for high-quality data and the scrutiny applied to its use, the organizations best placed to benefit will be those that can demonstrate not simply that they possess data, but that they understand what it is, where it came from, what rights attach to it and how it is governed. Providing greater certainty around those questions is ultimately what the Data Asset Foundations framework is seeking to achieve.

We've featured the best cloud computing provider.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

If the rumors are true, next week could mark the introduction of the first-ever touchscreen MacBooks. That's a Mac laptop with a touchscreen. People will probably embrace them and, no, I don't think their arms will fall off.

Steve Jobs famously hated the concept of a touchscreen MacBook. During an October 2010 product launch, he famously dismantled the notion:

"You know, the first thing you think about is this [he gestured toward the image of someone reading out to touch a MacBook screen]. Right, we thought about this years ago. We've done tons of user testing on this, and it turns out it doesn't work. Touch surfaces don't want to be vertical. It gives a great demo, but after a short period of time you start to fatigue, and after an extended period of time your arm wants to fall off; it doesn't work. It's ergonomically terrible. Touch surfaces want to be horizontal, hence touchpads for a notebook. That's why we've perfected our multi-touch trackpads over the years."

In short:

• It doesn't work
• Touch demands horizontal screens
• Your arm gets tired
• Your arm falls off

Jobs' comments came just 10 months after he unveiled the original iPad, which did have a sort of oddball keyboard accessory that put your hands way too close to the display, and two years before Microsoft introduced its first Surface Pro computers, which were basically tablets with kickstands and detachable keyboards.

Still, Jobs drew a line in the sand, and for more than a decade, Apple toed that line. Even when Apple seemed to split the difference with the Touch Bar on the 2016 MacBook Pro, Apple executives insisted a touchscreen was not the right move for Apple's popular line of MacBooks.

As former Global Head of Marketing Phil Schiller told Steve Levy back then,

"Our instincts were that it didn’t [make sense to do a touchscreen], but, what the heck, we could be wrong—so our teams worked on that for a number of times over the years. We’ve absolutely come away with the belief that it isn’t the right thing to do. Our instincts were correct.”

But unless the rumor mill is very, very wrong, Apple is set to prove Jobs and the rest of the old guard wrong and introduce its first touch-screen MacBook Pro next week in Los Angeles.

It will be, perhaps, a recognition that all those Windows OEMs who build 2-in-1 convertibles where the screen can fold all the way onto the bottom of the laptop or even detach to become a stand-alone tablet aren't wrong. There is a market for touchscreen computers.

What would Steve say?

What, though, would Steve Jobs make of this? The iconoclastic tech maverick, who died in 2011, would surely be appalled. Apple devotees might claim, "This never would've happened on his watch." Maybe they'll unfairly blame new CEO John Ternus, an actual engineer and 25-year company veteran who must've had a hand in the development of the new, alleged, touchscreen MacBook Pro. Surely Ternus was somehow involved over what was likely a 2-to-5-year development process, but the credit or blame should lie with now Executive Chairman Tim Cook, who probably greenlit the project.

If Steve Jobs were alive today, what might he say to Cook and Ternus about this? "Well, you better sell a line of arm sling accessories for all those aching muscles!"

No. In fact, I'm certain Steve Jobs would applaud the innovation. You see, Jobs was often adamant about something until he was not. He would insist Apple had no interest in making a phone, until the company delivered the iPhone.

There was a thing called the Steve Jobs "reality distortion field" for a reason. Steve Jobs didn't think an App Store and third-party apps were needed (or at least he claimed as much) before the App Store was unveiled. Video on the iPod was a bad idea until there was the technology to support it. Then it was a great one.

Steve Jobs believed in changing his mind. In an interview Jobs gave to Playboy as he turned 30, he said, "If you want to live your life in a creative way, as an artist, you have to not look back too much. You have to be willing to take whatever you’ve done and whoever you were and throw them away."

Jobs was more a marketer than engineer and, perhaps, more artist than tactician. In order for him to change his mind, the technology had to inspire him, or in order for him to be satisfied, he'd change the technology and specs (see the last-minute swap of the original iPhone's plastic screen cover to glass) until he was satisfied.

Sure, Jobs could say one thing on stage, and the company leadership would parrot it for years until, like Jobs, the innovation met the moment. And then, as Jobs was wont to do, change their minds.

So, if we do see our first touchscreen MacBook Pro next week, don't shed tears for Steve Jobs' discarded design principles. I'm sure he would say that this is as it should be and that a touchscreen MacBook Pro is, in fact, "magical."

Businesses invest heavily in attracting, developing and retaining talent. At the same time, many continue to view workplace technology primarily through the lens of procurement cycles, standardization and cost. Yet the technology employees use every day shapes how they experience the business and how effectively they can contribute.

As AI becomes embedded in everyday work, that connection becomes even more important. The organizations that realize the greatest value from AI will not simply introduce new tools. They will ensure employees have the technology, support and confidence to use them effectively.

The employee experience starts with technology

When workplace technology works well, it is largely invisible. When it does not, the effects are felt in everyday moments: a video call that starts late because something will not connect, a laptop that struggles through a day of meetings, or a workspace setup that is inconsistent from one location to the next. Individually, these are minor frustrations. Collectively, they influence productivity, engagement and perceptions of the workplace over time.

Employees do not judge technology by specifications or refresh cycles. Increasingly, they judge employers by the quality of the digital experience they provide and whether they can work without friction. As a result, organizations are increasingly looking beyond individual devices and considering the broader technology environment.

The PC, display, peripherals and collaboration tools all contribute to the employee experience, shaping how effectively people can collaborate, stay productive and adapt to changing business needs.

From access to enablement

This conversation is becoming more significant as AI moves from experimentation into everyday workflows. As these capabilities become more commonplace, organizations are recognizing that successful AI adoption depends on more than simply providing access to new tools. It requires the surrounding systems, processes and technology environments that enable employees to use them effectively.

This challenge is reflected in Microsoft's 2026 Work Trend Index, which highlights a growing "Transformation Paradox": employees are increasingly ready to work with AI tools, while many organizations are still adapting their systems, processes and operating models to fully support them.

While AI is often discussed in terms of applications and platforms, employees experience it through the devices they use every day. For most employees, the PC remains the primary workspace: the place where meetings happen, content is created, decisions are made and work gets done. As AI is embedded in everyday business applications, the role of the PC is evolving.

It is no longer simply a gateway to information or a communication tool. Increasingly, it is becoming a platform through which employees interact with AI-powered capabilities as part of their daily work.

As these capabilities become embedded into everyday workflows, employees can spend less time summarizing meetings, searching for information or completing repetitive administrative tasks and more time on work that requires judgement, creativity and human interaction. The outcome is not simply productivity in the traditional sense. It is a better day-to-day experience of work.

The technology behind workforce readiness

Workforce readiness increasingly depends on whether employees have the tools and support required to make effective use of emerging capabilities. This is where device strategy becomes increasingly important. Ageing technology rarely creates a single catastrophic failure. More often, it limits potential through delays, workarounds and interruptions that make it harder for employees to take full advantage of new ways of working.

At the same time, AI is raising expectations about what employees should be able to accomplish. If organizations want people to embrace new ways of working, they need to consider whether the technology environment supporting them is fit for that purpose. That means looking beyond refresh schedules alone and considering the broader role workplace technology plays in enabling productivity and workforce readiness.

Why device strategy is becoming talent strategy

Organizations do not build successful workplaces through technology alone. Yet it is becoming increasingly difficult to separate workforce outcomes from the technology environment that supports them.

That is why the conversation around AI PCs is expanding beyond IT and procurement to include HR and business leaders. As organizations look for ways to improve employee experience, strengthen productivity and prepare people for AI-enabled work, device strategy is becoming part of a broader talent conversation.

In the past, technology was viewed as infrastructure that supported work. In the AI era, it is increasingly becoming infrastructure that supports talent. Organizations that recognize technology as part of their talent infrastructure will be better positioned to attract, empower and retain the people they need to succeed.

We've featured the best AI chatbot for business.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Psychologist Abraham Maslow once warned: “It is tempting, if the only tool you have is a hammer, to treat everything as if it were a nail.” Agentic AI is becoming the enterprise version of that hammer.

Once people understand what agentic AI can do, the list of possible use cases grows quickly.

While that enthusiasm should be applauded, it can lead to strategies that are back to front “If AI is the answer, what’s the question?” By the time it comes to looking at results and whether the investment in AI has been fruitful, it’s hard to identify what’s been achieved and why achieving it was important in the first place.

Start with the value, not the agent

In the rush that followed ChatGPT, many organizations deployed AI tools to their workforces at large with no defined value or end goal. People started working differently almost immediately, but with no baseline captured, there was no way to know what had actually improved. Who knows how long it takes to write a coherent email and who knows what time is saved by having ChatGPT rephrase it for you!

This is what McKinsey calls the “gen AI paradox”. Its 2025 survey found just 39% of organizations reported any enterprise-level financial impact from their AI investment, despite mass adoption.

Organizations need to be precise about what they want to measure. There are three layers to this and most only capture the first.

• Adoption: are people using it?
• Behavior: are people doing their jobs differently?
• Outcome: revenue per salesperson, tickets closed per engineer, cycle time on a review gate.

Adoption and outcome together show a correlation; for example, people who use AI sell more. But it’s the, often missed, behavior layer in the middle that turns that correlation into causation, i.e. they sell more because they are now able to use AI to fill in more client reports, leaving more time to hit the phones.

Know your workflow

If organizations can’t see how work is really being done, two things happen. The baseline measures the wrong process, making any claimed improvement meaningless and an agent given the wrong steps cannot complete the job.

One procurement process at Atos looked like a four-step workflow on paper, but when we looked carefully and asked those doing it, it was closer to 10. People moved between Enterprise Resource Planning (ERP), Outlook and spreadsheets, copying information, waiting for replies and re-entering data manually.

A baseline on four steps would have measured the wrong process and an agent given four steps couldn’t have completed a job that takes ten.

Not every problem needs an agent

Understanding the workflow also shows whether agentic AI is the right technology at all. If tasks follow fixed rules “if this happens, do that” the answer may be Robotic Process Automation (RPA), workflow automation or a process change.

Agentic AI is better suited to workflows where the system needs to interpret context, coordinate across tools, make recommendations or act within defined boundaries. Even then, it needs a clear measure of success and a credible business case. The economics of deploying Agentic AI are still in infancy, but this will act as a natural filter for only the most valuable cases.

Prove the value before building

Value is where the tough work of a clear business case comes in. Proving the value of Agentifying a process before any code is written.

This requires access to good data sets. Evidence on cost and business value before introducing Agentic AI is needed before budget is committed. Where the benefits are measurable and enough open data exists, it’s possible to build something real in a week. Where the data still requires work, it is always worth refining the data you have before attempting to justify a business case.

Work managed upfront will save cost and time and give you evidence of value.

Build a portfolio, not a queue

Once there is a shortlist of use cases, organizations can prioritize against financial impact, feasibility, process impact, technology requirements, human impact and risk. Scoring across multiple factors rather than ranking on ROI alone creates a portfolio instead of a queue. A lower-value use case with fast impact might prove value early and build credibility for agents that follow. A high-value use case with a two-year runway may need to wait.

Budget for a new mortgage

Even once the right case has been selected, many organizations still struggle to see impact because they focus on the build, then under-resource the work needed to make the change land.

Gartner calls this a “transition mortgage”, where AI adoption demands more support than traditional technology, often requiring 25% more training effort and up to 200% additional change management.

In our own organization, we were guilty of building independent tools in different regions and teams to solve the same business problem. We have now established clearer oversight of our use cases. The issue was not the quality of the tools, but duplicated effort, cost and resources.

Not every workflow needs a hammer

C-suite leaders should start with fewer use cases where the value and workflows are clear, the baseline is measurable, the data is available and the role of the agent is well defined. Then they should budget for communications, adoption, measurement and scale.

The value from agentic AI is not measured by the number of agents an organization deploys. It is measured by whether the work changed and whether the organization can prove it. Agentic AI is a powerful hammer, but success lies in understanding which problems are nails and which are not.

We've featured the best AI website builder.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The most fascinating thing about Apple's new iPhone Duo – and others like it – may not be that it folds. It may be the innovation required to keep it cool. Because at a time when performance demands from edge AI continue to increase, we’ve now taken available smartphone real estate and – though not exactly halved it – constrained it more than ever. And that impacts heat. Which impacts performance.

What makes the Duo interesting isn't simply the engineering required to create a folding phone. It's that Apple is confronting the same challenge every edge device manufacturer now faces: AI capability is increasing faster than our ability to manage the heat it generates.

That’s why I believe we’re entering a new phase of device innovation, with thermal design increasingly shaping what’s possible. In the edge AI era, managing heat helps determine what a device can do, how long it can sustain performance, and how small or thin it can be.

The iPhone Duo is just the latest reminder that smartphone manufacturers have run into a stubborn engineering problem: How to keep mobile processors cool when there’s almost nowhere left to put a cooling system.

For its part, with the Duo, Apple has packed its A20 Pro and other heat-generating electronics into an exceptionally thin, foldable design. It’s also created a custom vapor chamber to help dissipate heat and sustain performance in the new form factor.

The latter is more than an engineering footnote. It’s a glimpse at a challenge the edge AI industry is starting to confront. Thermal design and industrial design are inseparable. And new solutions to thermal challenges must be considered.

Foldables Make the Challenge Obvious

Foldable phones make the thermal management challenge easy to perceive. Conventional smartphone design gives engineers a relatively predictable enclosure in which to arrange the processor, battery, cameras, and thermal management system. A foldable must accommodate a hinge and two display sections while remaining thin enough to fold comfortably and fit in a pocket.

At the same time, the processor and other electronics still generate heat – more, even, when we consider how people are starting to use mobile devices. Which means foldable designs necessarily concentrate significant heat-generating components into smaller and thinner portions of the device, precisely where space for thermal hardware is already constrained.

In general, vapor chambers remain one of the most effective ways to manage heat in a compact electronic device. They work by moving heat away from a concentrated source, such as a mobile processor, and spreading it across a larger area. But vapor chambers ultimately spread heat rather than remove it.

Sure, as processors become more powerful, manufacturers can increase the size and effectiveness of the vapor chamber. But eventually that thermal component itself becomes a design constraint, especially in foldable phones.

Agentic AI Makes the Challenge Urgent

In foldables, innovating around thermal management is itself an admirable design goal – if folding were the only technological advancement engineers had to tackle. But they’re pushing the envelope on physical design at the exact moment when smartphones are being asked to do much more. Namely, to process agentic AI workloads.

As agentic AI moves from the cloud to edge devices – AI-capable smartphones, wearables, glasses, and more – thermal management becomes the industry's most important challenge. Imagine a new foldable smartphone with its own specialized thermal management system, designed to accommodate an innovative new form factor. Now imagine asking that foldable to process vast amounts of data differently than before.

Unlike traditional mobile apps that perform discrete tasks, AI agents operate continuously. They sense their environment, preserve context, analyze streams of data, and act in real time.

In my view, the industry is still underestimating how significant this transition will be. Historically, a smartphone's thermal profile was defined by short bursts of activity: launching an app, streaming a video, capturing a photo. Agentic AI at the edge changes that model entirely. Devices are moving toward persistent intelligence, continuously sensing, processing, reasoning, and responding. With these sustained workloads, a device’s ability to manage heat becomes a defining constraint on its performance.

Passive thermal technologies remain indispensable. Vapor chambers, graphite, heat spreaders, and advanced materials will continue to play a critical role in future devices. The challenge is that Edge AI introduces thermal loads that increasingly benefit from more than heat spreading alone. As devices become thinner and workloads become more persistent, designers will need additional ways to not just to move heat away from sensitive components, but to move heat out of the system entirely.

Micro-Cooling Complements the Vapor Chamber

So here we are, with foldable smartphones capable of agentic AI processing. It’s a perfect storm of thermal design challenges. Engineers can continue enlarging or customizing the thermal systems they integrate into new and different form factors. Or they can find other ways to extract more thermal performance from available space.

The latter approach is where active, micro-cooling technology becomes a viable and important solution.

Unlike miniature rotary fans, which exist today and take up precious real estate of their own, MEMS-based piezoelectric actuators, fabricated in silicon, generate airflow inside extremely compact packages. Basically, it’s a fan on a chip, capable of precise active cooling that fits where conventional solutions cannot.

With micro cooling (µCooling), heat isn’t just spread out, it’s removed from the device entirely before it can sink into the case, delaying or eliminating thermal throttling. This hybrid approach to cooling in highly constrained spaces delivers better thermal performance in today’s AI-enabled smart devices.

It also creates greater design flexibility – a must as devices themselves become more physically flexible. If active µCooling can provide additional thermal performance, manufacturers may not need to maximize the physical dimensions of vapor chambers. A more appropriately sized, passive thermal system could be paired with localized active µCooling and thereby free up internal volume for other components and give designers more flexibility in how a device is assembled.

Put simply, with µCooling, thermal hardware wouldn’t have to consume as much of a device’s design envelope.

µCooling As Part of the Compute Architecture

The implications extend beyond folding smartphones. AI glasses, headphones, watches, robots, and other edge devices face the same fundamental problem, often with even tighter constraints on size, weight, and power.

The industry wants to put more intelligence into mobile devices while simultaneously making them smaller and more comfortable. That creates a physical contradiction. More compute generates more heat. Smaller devices provide less room to manage it.

µCooling offers a way to change the equation. Not by eliminating vapor chambers and other passive solutions, but by supplementing them with localized active cooling that provides additional thermal headroom while giving product designers more freedom.

As AI runs more continuously in the background, edge devices will spend more time computing and less time idle. Sustaining performance will matter as much as reaching peak capability. Foldable phones make this challenge acute, but it extends across the edge AI market as more intelligence and local processing place greater demands on compact hardware.

That intelligence has a thermal cost. Over the next decade, progress will depend as much on what physical devices can sustain as on what AI models can do. That’s why I believe thermal management is becoming a foundational technology of the edge AI era. The companies that solve this challenge will help define the next generation of intelligent devices.

Manufacturers already know this, which is why thermal management is so essential to product design. µCooling is the innovation behind the innovation – like smartphones that fold.

Check out our list list of the best product management software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

The "enterprise brain" is a term featuring more and more prominently in the enterprise technology market and at the C-suite level. Technology vendors of all varieties are scrambling to connect internal documents, knowledge bases and structured data systems into a unified architecture.

The compelling goal is to provide AI agents with as deep organizational context as possible so they can reason, automate and assist with scarcely believable accuracy.

But as businesses pour resources into linking their databases, policy documents and CRM entries, a significant blind spot remains.

Efforts to build an enterprise brain often overlook the intelligence contained in the millions of conversations employees are already having with customers, colleagues and candidates across phone calls and messaging channels.

Documents tell you what your policies are and CRM records can tell you what happened, but business conversations tell you why an incident happened.

For any company that is serious about building a true enterprise brain, conversations cannot remain a missing layer of organizational intelligence.

Beyond the confines of structured data

Standard business tools are excellent at tracking outcomes but terrible at preserving nuances. Structured systems may record instances of an enterprise deal stalled, an account at risk of churning or a candidate withdrawing from an interview process. But it of course can’t put its finger on the actual trigger.

A deal doesn’t just stall, there will always be a key factor that changed the buyer's mind during a late-stage call. Similarly a candidate doesn’t just decline an offer, maybe a competitor introduced a work-from-home policy during a quick negotiation chat.

When organizations rely solely on static databases to train AI models or inform strategy, they are feeding their enterprise brain an incomplete diet. They receive the post-mortem without the symptoms, and AI needs the sometimes messy, real-time context found in human dialogue to understand the operational reality of a company.

Actionable context

Artificial intelligence has reached a level of maturity where analysing high-volume voice and chat interactions is no longer a nice to have or a bonus. But just recording every interaction and dumping raw transcripts into a data lake, which effectively hoards unstructured audio, solves absolutely nothing. To make conversational context truly useful, businesses must focus on a few core operational pillars.

First, conversational insights must automatically sync with existing enterprise systems rather than living solely within a VoIP dashboard, allowing real-time data from a support call to enrich the CRM and trigger automated alerts. Second, the extracted intelligence needs to be routed immediately to the teams equipped to act on it. For example, if sales calls routinely show that prospect friction stems from a specific pricing tier, that signal should flow directly to product strategy and marketing rather than remaining trapped in the revenue organization.

Finally, while AI should perform the heavy lifting of summarization, sentiment analysis and pattern recognition, human judgment must always remain the final filter for important decision-making. While this may seem obvious, it is amazing how overly reliant on AI many employees have quickly become, including using it for critical business judgement calls.

Breaking down knowledge silos

Knowledge silos continue to be huge challenges for many organizations. Sales teams learn what buyers want, support teams learn where the product fails, and recruitment learn how the market perceives employer branding. All is highly valuable information, but all too often these groups don’t share a unified intelligence pool.

When conversational context is ingested into the enterprise brain, these walls dissolve across every department. For instance, when a prospect mentions a niche feature gap during a sales demo, that insight can prompt product teams to prioritize the feature while sales updates its outreach messaging.

Similarly, if a customer casually mentions an upcoming corporate restructuring on a support call, account management can initiate proactive outreach to protect the account before contract renewal.

By connecting these conversational dots across departments, organizations shift from reactive firefighting to proactive alignment.

Value over volume

Now this is not to say that trying to capture, transcribe, and analyze every single internal interaction is the way forward. Because it goes without saying, this is borderline impossible. If attempted, it can lead to sky-high compute costs and even legitimate anxiety regarding employee monitoring.

To build a functional, trusted enterprise brain, organizations must prioritize value over volume. The focus should be on high-value workflows where conversation directly drives business outcomes, such as customer-facing revenue calls, critical support interactions or key talent acquisition touchpoints.

Focusing on these targeted areas minimizes data noise while delivering immediate ROI. Organizations must build trust by maintaining clear boundaries. Conversational analysis should focus on market trends, product feedback and operational blockers instead of micromanaging employee tone or enforcing rigid scripts.

The complete enterprise brain

A truly intelligent enterprise cannot run on paperwork and revenue tables alone. It requires the context and nuance that only human interactions provide.

As enterprise AI continues to evolve at rapid pace, having the largest static document libraries will count for nothing. The organizations that thrive will be the ones who successfully pair their structured business data with the rich, dynamic intelligence hidden inside their everyday conversations.

Checkout our list of the best business intelligence platforms.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

AI development and expansion is rocketing across our society, sparkling and bright like a Roman candle, but with its tail on fire. The disruptions are coming so fast, and the concerns are rising so quickly, that no one, not even those running the biggest AI companies, knows what to do. Instead of addressing issues like AI slop, job loss, self-replicating and developing AI running amok, and data centers, these AI bosses have come up with a better solution: cold, hard cash.

On Monday, Elon Musk casually posted on his X platform that he would give one of his SpaceX shares to every child in America, with the promise of more to come. Over at Anthropic, Dario Amodei proposed a public fund built on the value of 5% of his company's shares when it goes public next month.

Of the two offers, Anthropic has the most obvious value. The company might be worth a trillion dollars when it goes public in November, which means the fund will be close to $50 billion. That's a lot of money.

Whether it's $150 (the approximate value of a SpaceX share) or $5 billion, neither offer actually addresses AI's myriad issues. None of this equals regulation, oversight, a pause, or a concrete global discussion about a technology that may in 12 months' time be able to do any of our jobs, access most of our systems, and start making decisions on our behalf. None address the real-time, growing cybersecurity threats posed by AI that can crack any encryption and set off Denial of Service attacks at unimaginable scales.

But how?

Let’s leave aside all that for a moment and talk about the nuts and bolts of these proposals.

First of all, there’s the question of how Elon Musk is going to give every one of the 72 million children in America a stock grant (assuming — a big assumption — he follows through). There are ways of doing this (custodial accounts, transfer to a brokerage, even platforms like GiveaShare that support stock gifts for children). Trump Account recently did something similar (sort of tiny Individual Retirement Accounts for pint-sized people). However, when those stocks go in there, they're not something the children will access now or in the near future.

What would be the value of $150? A couple of toys, a few outfits? More likely, what’s going to happen here is those SpaceX shares will sit and maybe accumulate value (maybe not), and then, when the child is an adult looking for one of the few jobs AI has not taken over, they’ll use it to help support themselves for a little while. Maybe buy a meal.

As for Amodei's Anthropic offer, well, it is a lot of money, but once you set up the fund, you’d figure out how to manage it and allocate it. Who is going to control that fund, and how will it be parceled out? Will it, for instance, go to every individual state? Will it go to schools, grade schools, maybe high schools, maybe universities? Will it be used to set up AI curriculum? If so, what curriculum built now will be useful a decade from now? Those billions could be used, but they’ll have to be sort of churned and replenished so that new curriculum could be built (probably by AI).

I’m not saying that’s what’s going to happen because it could just be for job training centers, retraining for new jobs that people have never done before, but that AI cannot reasonably do. Maybe Anthropic will be funding trade schools. I have yet to see an AI plumber or electrician, though I guess eventually the robots will be skilled enough to run a pipe or fix a circuit.

Of the two plans, I have far less confidence in Musk to hand over those 72 million shares because he’s made promises like this before and has not carried through. It becomes a bit of a show, a flurry of hands and a digital wink to convince people that he actually cares, but there’s no indication that he really does beyond the growth of his company and personal wealth.

Money for something

Now, these things are not mutually exclusive. I do believe we need money to help retrain vast workforces in this AI age, and that should happen sooner as opposed to later. And sure, you know, giving a child a share is not a terrible gesture, a little something to put aside for later, for a rainy day, or a day when they realize that the AI is in control and the only way they can buy food is directly from another person because the AI is running the grocery store.

Maybe though, Anthropic's Fund could be used to figure out a better, more environmentally and community-friendly way to build data centres, perhaps in places where both people and wildlife don’t need to live (space, underground). I don’t know what the answer is to that, but that research needs to happen.

It’s unclear if either one of these monetary plans will actually come through, but one thing I can guarantee is that AI is not slowing down. Yes, there’s probably a coming correction, and there may be more consolidation, but the trajectory is set. It will continue, and if we don’t get people together to actually talk about how we’re managing this revolution, no amount of money is going to solve our AI problems.

Amazon has found another way to make itself part of our everyday lives with its new Bee AI devices. The Bees can record conversations and are pitched as a way to remember important details and translate plans into actions. Naturally, those actions can include buying things from Amazon, making every casual conversation a potential shopping cart. You can pay $75 or $100 for the privilege, depending on whether you choose the Bee Frontier, which clips to a wrist strap or clothes, or the Bee Eclipse, a magnetic smartphone attachment.

There are practical applications for something that remembers conversation details you might otherwise forget. An assistant that recalls an appointment, prepares a follow-up email, or reminds you about a promise could save considerable time. But Amazon's announcement also promises personalized shopping recommendations based on what Bee learns about you, which strikes me as a depressing, if inevitable destination for AI tools.

Of course, you might argue plenty of people have invited Amazon into their homes through Alexa devices. Plus, Amazon already has a potentially huge amount of customer data from shopping histories, which lets it recommend products so well that it's a cliché joke to feel like you only have to mention a topic out loud for related items to show up in your recommendations.

That joke is an explicit feature of the new Bees. Turning spoken conversations into another source of personalized assistance is supposed to appeal to potential customers. The question is whether this is a bridge too far when it comes to allowing Amazon to eavesdrop on our lives to shape our shopping.

Idle chat becomes a shopping list

The idea of an AI assistant remembering what you say and not letting anything important slip through the cracks is enticing for the convenience it promises. But imagine casually mentioning that your back hurts, your car is making a strange noise, or your child has developed an interest in dinosaurs. Those aren't just setups for a reminder to make a doctor's appointment, bring in your car, and buy tickets to the museum you'll get. They're also data points for potential purchasing interests that might otherwise never have reached a retailer.

Amazon hasn't said Bee will automatically turn every overheard complaint into a product recommendation, yet the shopping integration creates an obvious path to do so. But it might be too much personalization. Adding information from spoken conversations to what Amazon already knows could paint a much more intimate picture of someone's life, good and bad, regardless of Amazon's promises to keep data under the user's control.

And there are real privacy protections in the system, according to the company. Bee's AI processes audio in real time and deletes it rather than retaining recordings, while conversation transcripts are encrypted and users can delete them. The devices also include physical controls and lights that indicate when they're listening. Plus, the Amazon shopping connection is optional and opt-in.

But deleting the original audio doesn't eliminate every privacy concern when a device is specifically designed to remember and interpret what people say. The information extracted from a conversation can be just as revealing as the recording itself.

Eavesdropping Bees

The privacy problem becomes even more complicated when other people enter the conversation. Someone wearing a Bee device might be perfectly comfortable having their conversations analyzed, but their friends and colleagues haven't necessarily agreed to participate. A small recording indicator is hardly a substitute for meaningful consent, especially when people may not recognize the device or understand what it does.

We've already seen this controversy unfold with Meta's Ray-Ban smart glasses. People have reported unexpected recordings, and even filming. Bee doesn't have the same camera-related concerns, but recording conversations introduces its own problems. A friend discussing a medical diagnosis or workplace dispute in confidence might not feel so open if they know you're wearing a Bee. They haven't necessarily agreed to have an AI system connected to the world's largest online retailer convert their comments into searchable data.

There are also practical questions about where recording is appropriate. Private conversations can involve confidential information, and recording laws can be wildly different depending on location and circumstances. Anyone using Bee needs to consider the consent of the people talking around them, regardless of whether Amazon deletes the original audio.

The fact that personalized shopping is part of the product's ambitions says something revealing about how the company imagines the future of personal AI. Amazon and Meta seem convinced that the future involves devices observing more of our lives so they can anticipate what we need next. And it's easy to think of reasons to develop these capabilities. People with memory difficulties, busy professionals, and anyone struggling to manage complicated schedules could benefit from reliable reminders and searchable conversations. But making recording devices an ordinary accessory could have far-reaching consequences, not all pleasant.

For me, the shopping integration is where Amazon's vision becomes particularly unappealing. I can appreciate an assistant remembering that I promised to call someone or reminding me about an appointment. I have considerably less enthusiasm for a device that might turn a casual remark about needing new shoes into another opportunity for Amazon to suggest something I should purchase.

We've spent years teaching technology companies about our interests through searches, purchases, clicks, and browsing histories. Wearable AI assistants like the Bee could extend that relationship into our conversations with other people. I don't particularly want my conversations becoming another source of shopping recommendations, though. If it's that important, maybe just write it down.

• The Bing Wallpaper app has seen an annoying feature reintroduced
• This means those clicking on a blank space of the Windows 11 desktop will get a Bing search fired up in their default browser
• Unlike the last time Microsoft tried this stunt, there's no way to disable the ability this time around

Microsoft is up to its old tricks again, except even worse this time — with a scheme whereby Windows 11 users with the Bing Wallpaper app installed are sent to its search engine with a mere stray click on the desktop.

Windows Latest reports that an old, potentially off-putting shortcut has been reintroduced to the Bing Wallpaper app, whereby if you click anywhere on the Windows 11 desktop, a search in the (separate) Bing engine is fired up (in your default web browser).

If you recall, this actually happened a year ago, but Microsoft reversed course, although now appears to be trying to reintroduce this feature. The difference is that this time around, Windows Latest notes that there's no option to turn off the functionality (which was present last year, in the app's settings).

However, there are some points to note here: firstly that you must choose to install the Bing Wallpaper app (which cycles Bing images daily for your desktop background). It's not a default Windows 11 app, so this isn't something Microsoft is piping directly to Windows users.

Secondly, the Bing search doesn't trigger every time you click on a blank piece of desktop (a space without a file, folder or icon). It does this once, and then the Bing search won't be triggered for another 24 hours. You'll only experience this once per day, in other words.

Analysis: Microsoft must do better

Just because this app isn't part of Windows 11 by default doesn't give Microsoft the excuse to push functionality like this. It's just going to annoy or even confuse people if they accidentally click the desktop and end up with a Bing search being loaded.

The search relates to the wallpaper image, in case you're curious, and fair enough – maybe some folks might be inquisitive and actually want to know more about the picture by clicking on it.

However, this should be an ability that's optional and can be turned on, not something that's enabled by default. What's worse still is that in this new incarnation of the click-to-Bing farce, you can't even turn it off if you want to do so.

Why is Microsoft engaging with this kind of nonsense to promote its own services? I really can't begin to guess given that 2026 is the year of fixing Windows 11 and curing all the ills around the OS, and while this app may not be part of the operating system, it's still a piece of official Microsoft software – and there shouldn't be any place for this cynical Bing traffic grab or any other remotely anti-consumer practices.

Whether it's a part of Windows by default, or not – or whether you have to be 'foolish' to install the Bing Wallpaper app, as some online commenters argue – is besides the point. Microsoft is supposed to be better than this now, but another episode with a recent Harry Potter advert in this Bing app – one that provoked a considerable rebellion – suggests that it very much needs to try harder.

Organizations can be under huge pressure to make changes to their IT estates. In particular, there are two drivers of this change.

Firstly, there’s the pressure to innovate. The rapid advancement of technology and AI coupled with shifting business priorities can put organizations under pressure to adopt the latest tools, with vendor messaging advocating for the speed of adoption. Simultaneously, ongoing inflation and slower economic growth can mean there is a need to find more cost-effective solutions.

Secondly, there’s the change that is forced. Operating systems might be reaching their end of life and need upgrading, or there are outages and cyber threats to defend against. There have also been several recent examples of software vendor prices rising significantly due to ownership changes, like with VMware, and companies therefore needing to find new solutions for their budget and reduce vendor dependency.

However, in dealing with these pressures, the trap companies can fall into is trying to deliver all of this change at once and within a small timeframe. Yet in doing so, this change becomes harder to manage and the risk of outages and system failures increases. So, how can they deliver any change that is needed safely?

The dangers of too much change at once

The main challenge in changing too much too quickly is visibility. Companies have to oversee complex IT estates – on-prem systems, the cloud, servers, end-user computing, networks – where change is constant and security updates are largely automated. As this change happens, there are more interactions between older and newer systems, different operating systems, different products, and this means the risk of failure can increase.

Configuration drift, for example, occurs when changes, which haven’t been recorded or authorized, cause a system to drift away from its intended baseline (of how it should perform). Without the right monitoring software, this drift can remain invisible, creating added risk to every subsequent change – and that’s when an outage or cyber incident happens.

In May, for example, Google Cloud suspended Railway’s production account without warning because Railway couldn’t see that the platform had a technical dependency on GCP that caused it to fail, leading to an eight hour outage.

What’s more, as we’ve seen with the adoption of AI tools, change can happen faster than training and processes can keep up with, further increasing the risk of disjointed workflows, security vulnerabilities and data breaches. But even when teams can see the changes that need to take place, they can also be faced with the challenge of deciding how to modernize their existing applications.

Finding alternatives to the buy-or-rebuild dilemma

Of course, trying to slow down change isn’t always possible. A company might have been priced out of its VM software or needs to upgrade its applications for security purposes. In this position, traditionally, IT teams have had to either buy new software or rebuild existing apps, which can both be expensive and time-consuming.

This can lead to vendor lock-in, as IT doesn’t see a viable alternative and doesn’t have the engineering capacity to rewrite applications tied to their original operating systems.

To overcome this dependency, as agentic coding matures, organizations may increasingly opt for a third option: using AI agents to build their own replacements for SaaS tools. But while this can appear to remove vendor lock-in and license costs, it also transfers responsibility for maintenance, security, configuration and ongoing support in-house, making it even harder to oversee change without additional resources or capacity.

However, there is another route. Packaging and migrating existing applications can allow teams to modernize without taking on the burden of building from scratch. By packaging an app’s or virtual machine’s existing operating system dependencies, teams can move them into containers and modern desktop infrastructure without needing to rewrite the app.

The app maintains its existing behavior, but can now work on the latest operating systems and in workspaces like AWS.

Delivering change at scale

With a modernization strategy formed, how do teams gain ongoing visibility across their entire estate and know when apps need to be packaged?

With so much change taking place, manual monitoring and reconciliation processes will struggle to keep up with the demands of infrastructure modernization. Instead, teams should consider using configuration monitoring software to gain continuous visibility into change across their entire estate.

Drift detection tools, for example, can automatically measure change across an estate against established configuration baselines and alert IT teams whenever a change triggers a system to move away from its baseline. This means that an automated patch by an external provider or a manual intervention by a user gets detected and flagged to IT, changes that might otherwise go unnoticed and trigger an outage further down the line.

Imperatively, this estate-wide visibility means teams can see and understand the impact of any unauthorized changes taking place and intervene before they lead to failure.

It also gives them the information they need to understand how a transformation project might interact with the systems currently running in their estate, helping to inform the feasibility and implementation of the proposed changes and deliver more targeted, impactful and safer change.

Alongside configuration visibility, it’s also possible for teams to automate the packaging and delivery of apps, saving them a large amount of time and cost in manually performing a typically laborious task. This subsequently creates a repeatable and scalable process that enables apps to be regularly deployed and made available to users.

Escaping the trap

With so much pressure to change, above all, escaping the transformation trap requires a mindset shift. It involves, one, seeing change as an ongoing problem to manage, and two, not attempting to carry out too many innovation projects at once.

Small, incremental changes can deliver more meaningful impact and ensure systems are properly deployed or integrated, while this targeted approach also provides more capacity for IT teams to deal with any forced change that might surface.

Yet by prioritizing change visibility, IT teams can mitigate forced change by addressing risks earlier and seeing when upgrades are needed in good time. And by automating processes like packaging apps, they gain more capacity to prioritize innovation projects as routine changes take place faster and more securely.

When these processes are joined together, instead of grappling with an overwhelming and uncontrolled level of change that increases risk, teams can ensure change remains a driver of security, resilience and innovation.

We've featured the best IT automation software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

There has been an explosion of AI-powered products released over the past four years, with most built on top of leading foundation models. Startups, enterprise software vendors, and even large incumbents route their core intelligence through APIs from the leading AI labs, and a growing but still narrow set of open-weight alternatives.

This arrangement has most certainly enabled a genuine burst of AI products, but it has also created a structural fragility which can be remedied through the democratization of model creation and training.

Defining a wrapper

In its simplest form, an AI wrapper is a product whose core function is an interface sitting on top of a third-party foundation model's API, with little to no proprietary technology beneath that layer.

That approach can have a short lifespan and Jasper can illustrate that. Once the standout of the AI writing boom with a $1.5 billion peak valuation, Jasper saw its core market vaporized when ChatGPT introduced free, native writing features. Its revenue fell from approximately $120 million in 2023 to approximately $55 million in 2024. Jasper did survive, by redirecting its efforts toward enterprise workflows and brand-specific training.

Generally speaking, analysis shows that, among AI startups that have wound down in the US in 2025, 81% were "wrappers and apps". Copilots and assistants layered on foundation models, AI productivity tools, content generators, and vertical SaaS products whose core differentiation was an LLM front-end rather than proprietary data or infrastructure.

In Jasper’s case, and in many others, what pressured the company wasn't a competitor out-executing them. It was the platform they built on top of simply absorbing the feature they'd built their company around.

Not all wrappers are equal

It would be a gross generalization to say that all AI wrappers are the same. A thin prompt-and-display wrapper is different from a product that does heavy retrieval, tool orchestration, agentic workflows, or domain-specific fine-tuning. The latter can have real defensibility even while sitting on top of a third-party base model. The model call is one component, but certainly not the whole product.

Some companies also use multiple foundation models or fine-tune/distill their own smaller models for specific tasks after prototyping on a frontier API. Open-weight models add variety too as not all wrappers call a closed-source frontier model.

The wrapper problem

Regardless of where an AI product sits on the wrapper complexity scale, they all share the same problem. When a company's core value proposition depends entirely on an API call to a third party, that company does not fully control its own product.

It cannot control pricing changes, rate limits, deprecations, shifts in model behavior, or the possibility that the platform provider decides to compete directly in the same vertical.

This is not a theoretical risk, though how much it affects any given company likely depends on how much of its value sits above the model layer.

The economics of the wrapper

So what is the defensibility of an AI wrapper in practice?

The short answer is that it comes down to intellectual property.

Where a company benefits from an existing distribution advantage, established customer relationships, regulatory licensing, network effects, or years of accumulated brand trust, model-layer differentiation may not matter as much. But if what makes a wrapper distinctive is observable from outside the organization and replicable by a competent competitor within a limited period, it seems reasonable to expect it will eventually be replicated.

That is the unfortunate economic truth beneath the wrapper model. The cost of replication is near zero for the platform that controls the underlying model, and near total for the startup that doesn't. A wrapper's moat, if it has one, has to come from something other than the model call itself. Absent those, a wrapper is unlikely to have a lasting competitive advantage.

The barrier to the solution

At this point, you might be inclined to agree that model ownership is one solution to this problem. But that's easier said than done.

The reason so few companies own their models is not a lack of desire. It’s, more plausibly, financial. Training a frontier-competitive model requires enormous compute budgets, specialized talent, proprietary or expensively licensed data pipelines, and the infrastructure to run reinforcement learning at scale. These costs put model creation out of reach for all but the best-funded companies.

For companies to build on their own models as a norm, this barrier has to fall meaningfully. Several forces are already pushing in this direction, open-weight models being one of them. But even then, hosting and fine-tuning an open-weight model is no small task. It still demands technical talent and compute infrastructure most startups don't have access to.

True democratization of model ownership

Democratization, in this context, doesn't mean every company training a frontier-scale model from scratch. That goal is neither realistic nor necessary. It means democratization reaching the training layer the same way access to AI has been democratized for end users.

Democratizing model training and ownership could be what gives founders a lasting chance at turning an idea into a business that doesn't eventually get absorbed by a model provider as a simple feature. And if that happens, the stakes may go beyond individual business durability.

As things stand, access to models is cheap. Access to training is not. This needs to change. The ability to train an AI model should be fully democratized so that the value, the control, and the responsibility can be distributed rather than centralized.

We've featured the best AI tool.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

You don't need to understand the mechanics of AI tools to understand what's happening to traditional technology pricing. AI is breaking the link between effort, cost and value that technology services have relied on for decades.

Imagine two managed service providers responding to the same production incident. One assigns an engineer who spends five hours tracing logs, checking dependencies and isolating the root cause.

The other uses AI-enabled observability to identify the issue in minutes, trigger the appropriate remediation and restore service before users notice. The second provider used fewer hours but delivered more value. Should they be paid less?

Enterprise technology services are running into the same pricing challenge, but on a larger scale. The old pricing model equated more hours with higher cost and, in theory, greater value. But today, AI is enabling organizations to move from reactive support to self-healing operations. In our own client engagements, we're seeing auto-generated tickets fall by as much as 70%, roughly one in five operational events resolved autonomously, and up to 45% fewer major outages where predictive models are in place.

That should translate into straightforward cost savings, but the reality that CIOs and CFOs are grappling with is more complicated. AI can deliver the same or greater value, with significantly less effort, which undercuts the assumptions behind what they are paying for. Managed services providers are facing the inverse of this same problem: How do you price services when the traditional relationship between effort, cost and value no longer holds?

Increasingly, predictability is replacing capability as the bigger challenge to address.

What senior leaders are saying about AI

The challenges of identifying the right AI use cases are coming up in conversation after conversation with CXOs and increasingly with business leaders beyond IT. Leaders of all kinds are recognizing that a successful AI program begins with an operational problem, and the sophistication of the AI needs to be matched to the scale of that problem.

That also means organizations must be willing to redesign their operations around what AI makes possible. Only then can clients and providers have a meaningful conversation about value and how to price it. For example, one enterprise client's finance team was struggling to get a single, accurate view of its cash position across dozens of markets and currencies.

Each region had recorded the same transactions differently for years, and leadership had spent more than five years trying to standardize the process without success. Before this, there was no credible baseline for measuring improvement, making it nearly impossible to agree on the value AI was expected to create.

A client’s business operations are often just as unsettled as the technology helping run them. The contracts holding the two together were built for a past world with more predictable staffing, fixed scope and stable demand.

There is no one right pricing model

The natural response to this uncertainty is to look for a new pricing model to replace the old one. But no single model solves the problem. Outcome-based, consumption-based, and AI-enabled pricing approaches each have advantages in the right circumstances and introduce new trade-offs. The challenge lies in creating enough flexibility to operate when effort, cost and demand no longer move predictably.

Outcome-based pricing works well when a result can be clearly defined and attributed. That gets much harder once AI-driven workflows start feeding into one another, making it difficult to trace which system or step contributed to a business result several automated stages downstream.

Consumption-based pricing ties cost to usage, which sounds fair until you run into the token paradox. While the price of a single AI query keeps falling, total spend keeps climbing anyway, because usage is no longer limited to one person typing one prompt. It's chains of AI agents running continuously in the background, each step triggering the next and including its own line item.

Take a lesson from a banking client running a 20-developer modernization program. Token costs per developer had climbed to roughly $1,000 a month. The team reduced that to $200 by routing simple work to lower-cost models, identifying usage patterns where token consumption was four times higher than the team average, and using deterministic automation for repetitive tasks where AI wasn't needed.

Gain-sharing models sound intriguing because the provider takes on more risk in exchange for a share of the outcome. However, they can become difficult to manage when both parties disagree about which improvements resulted from the provider’s work and how much credit each side deserves.

Why pricing flexibility is vital

These pricing models each assume either effort, cost or demand will remain predictable enough to price against. Increasingly, that is not the case.

While the market has not settled on a single approach, we believe hybrid structures have the greatest potential to become the preferred commercial model for AI-enabled managed services. A hybrid model combines a predictable base fee for core services, a variable component tied to AI consumption and outcome-based incentives where results can be accurately measured.

For CIOs, this approach provides greater budget predictability while offering some protection against consumption spikes. For providers, it helps manage AI consumption volatility while creating incentives for efficiency, continuous improvement and value creation.

Designing partnerships for uncertainty

The goal was never to find the perfect pricing model for AI, but rather to build commercial structures that can flex when conditions change again. That starts with determining clear baselines before any gain-share or outcome-based conversation begins.

It continues by intentionally building in scheduled points to recalibrate scope and pricing instead of being locked into the traditional three-year contract. Transparency around performance, costs, governance and AI outcomes also needs to be prioritized so that both sides trust the numbers.

We are still in the early innings of figuring this out. But it’s clear that AI is changing the role of the technology partner, from delivering capacity to helping enterprises manage uncertainty and accelerate business outcomes. Enterprises must build enough flexibility into their pricing models to adapt as the technology, demand and value equation changes.

While no one can perfectly price the future, we can design commercial models that evolve alongside the technology and the value it creates.

We've listed the best accounting software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit