An investigation into the alleged misuse of AI at Derbyshire Constabulary, arriving alongside wider scrutiny of generative tools across UK policing, has raised an uncomfortable question that extends far beyond the force.
What happens when powerful technology enters any organization where sensitive information is used to support decisions that may later need to be explained and defended?
The answer is not to keep AI out of investigations, intelligence or other high-stakes work, but to deploy it in ways that can withstand scrutiny. Within secure, auditable systems, with clear boundaries and human oversight, AI tools can help organizations connect information, manage growing caseloads and identify important patterns while ensuring that every insight can be traced, explained and defended.
The risk arises when employees turn to general-purpose tools that were never designed to meet those evidential and security standards.
Co Founder and Chief Strategy Officer at Clue Software.
The pressures that make AI appealing to teams handling high quantities of sensitive data will be familiar across public and private sectors. Caseloads grow, data sits across disconnected teams of investigators, analysts, HR professionals, safeguarding leads, legal, security and senior decision-makers, and a tool that can condense lengthy case reports in seconds is difficult to ignore.
The trouble begins when that tool has never been approved and sits outside the security perimeter, leaving no record of what it was asked or what it returned. A potentially more serious risk is that, in its pursuit of brevity or its interpretation of the facts, it may omit or overlook information that is vital to an investigation. When guilt, innocence and ultimately lives are at stake, such omissions are unacceptable.
At that point, the risk reaches into evidential integrity, regulatory compliance and, in policing, the administration of justice.
This is a governance problem that can surface anywhere sensitive and high-consequence information is being interpreted, summarized or used to support high-stakes decisions. The more useful AI becomes, the more important it is to define the conditions under which it can be safely used.
Any organization using AI to handle this type of information needs a policy that states which tools are sanctioned, what they may be used for, and where human judgement remains non-negotiable. Where those boundaries are left vague, people will draw their own, usually under deadline pressure.
Where generalist AI falls short of evidential standards
Intelligence and investigations are built on provenance and auditability. Every conclusion should be traceable to its source, and every step should withstand challenge months or years later.
Most generalist AI tools were never designed to carry that burden. Feeding sensitive intelligence into a third-party platform raises immediate questions about security, compliance and data sovereignty. Once material leaves a controlled environment, an organization loses sight of where it is held, who can reach it and whether it is being retained for model training.
Then there is the question of accuracy. Generative systems are constantly improving, but there is still margin for error. For example, the most sophisticated models are achieving hallucination rates as low as 4.2%, down from 12.2% by previous versions. This demonstrates that these tools are becoming more reliable with maturity, but there still needs to be scrutiny.
AI generated outputs cannot be fully trusted at face value, they need to be rechecked and validated. While presented as fact, information can be delivered outside of appropriate context or can occasionally be hallucinated analysis. These can pass into a report before anyone thinks to test it.
When AI-generated material is wrong or impossible to verify, the consequences run through everything that follows. Weak evidence undermines prosecutions, delays safeguarding decisions and hands opportunities back to the people we are trying to disrupt. None of this makes the case for keeping AI out of the process. It makes the case for keeping the wrong kind out.
What purpose-built investigative AI does differently
There is a place for AI within intelligence and investigation operations, however. Specialist platforms approach its use with auditability and evidential continuity designed into the foundations rather than bolted on later. AI accelerates the parts of the job that consume time, connecting the dots without adding judgement.
It can flag when the same entity – such as an individual, bank account or company - appears across intelligence reports, transaction records and internal correspondence, presenting those connections for an officer to assess. Patterns can surface in minutes, rather than through weeks of manual cross-referencing, and the decision about what it means stays where it belongs.
These platforms also operate inside secure environments, giving organizations control over how sensitive material is stored and processed. Queries, outputs and decisions can all be documented, which makes accountability far easier to demonstrate at the moment it is questioned.
Governance decides whether AI holds up
Trust depends on how a result was reached, just as much as the result itself. As AI becomes embedded in casework, organizations will be asked to show that it was used fairly, proportionately, and properly supervised. Technical controls alone will not answer that.
A governance framework is needed to effectively assign responsibility for AI-assisted decisions, set out how outputs are validated, and ensure the people using these tools understand where the limits sit.
It should also be explicit about where AI has no role. Some activities, particularly those bearing on legal proceedings, safeguarding or an individual’s rights, are not appropriate for automated assistance at all. Drawing that line matters as much as the choice of platform.
Innovation only works when it can be defended
Recent scrutiny of policing does not weaken the case for AI in intelligence and investigation operations. It shows where the failures originate, which is in unmanaged and unauthorized use, rather than in the technology itself.
The organizations that get the most from it will be those that deploy it inside secure, auditable environments where every insight can be traced, explained and defended. Integrity is what allows innovation to survive contact with a courtroom or a public inquiry.
We've featured the best data loss prevention.
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