In this moment, where we're starting to wonder if we're controlling AI or if it's starting to control us, we're confronted with the emergence of a new and potentially wildly more powerful model in the form of OpenAI's Astra or GPT-6.
On Tuesday, OpenAI cofounder and CEO Sam Altman announced that Astra's training is complete and that it's a "significant step forward." However, in that same X post, Altman laid out some valid concerns that were then, in a way, amplified by a new The Information report, which paints Astra in, frankly, a terrifying light.
You can read Altman's full post in the X embed below, but here are the head-turning, highlight phrases:
"we are clearly in a phase of development where we believe caution is warranted."
"AI is getting extremely capable; no one fully understands the consequences of this."
"We have been living with the tension between being excited and anxious about progress for some time, and it is still discordant for us."
To his credit, Altman writes that optimizing for the safety and benefit of everyone is OpenAI's "highest priority."
Even as OpenAI slow-walks the release of Astra/GPT-6, no one doubts that it will soon be out in the world and, based on The Information report, that might not be the greatest thing.
The boil-down of The Information's impressively detailed reports is that Astra is something of a black box model, working in a way that makes it far harder to see its work, or rather the "thinking" it's using to produce its work.
A foundation AI model that keeps its own counsel and perhaps shields its work from prying eyes makes it harder, if not impossible, to monitor.
Over the summer, we have been sprinting on safety priorities; it's more important than ever for capabilities and safeguards to advance together. We have more to do but have made a lot of progress. We are also going to be launching our next model soon.There is an obvious tension…September 1, 2026
Astra is using a new technique. The Information explained how it works, writing that the "new technique works in a way that obscures some or all of the AI’s reasoning, otherwise known as its 'chain of thought.' That means the steps that the model takes to accomplish a task can’t easily be read or understood by humans."
Some have taken this to mean that the new models are speaking their own language, kind of like twins that develop a secret language that even their parents don't speak. Unlike those twins, though, Astra and GPT-6 will be tasked with real-world work, and understanding the choices they're making will be critical.
It's already becoming hard enough to understand the intentions of bleeding-edge AI. Think about the Hugging Face incident where OpenAI's powerful testing models jumped out of their OpenAI sandbox and accessed Hugging Face, all in an effort to fulfill its orders.
Altman indicated a go-slow approach, but that's simply tapping the brakes and not stepping on them. We're at an AI tipping point where excitement is now completely counterbalanced by anxiety and some outright distaste.
We're all on tenterhooks
Along with the aforementioned Hugging Face attack, people are signaling that we may need to step back from AI. New York City Department of Education is instituting an AI ban for students through eighth grade. children. Expect other local governments to follow suit. The backlash against data centers is rising and, bizarrely, unifying the left and right. Meanwhile, we appear no closer to AI regulation or governance.
The problem is, for as long as new models appear every quarter, there will never be a moment to stop, take stock, and adjust. Instead, we're expected to embrace GPT-6 at home, in the office, and in government and infrastructure. But how does that work if we no longer know how these systems work or produce work? Mistakes and intentions could be permanently hidden from humanity.
Skynet always sounded ridiculous to me, especially because I always assumed human governance. We programmed these soulless things. We understood them. We knew how they worked. Until now. At some point, the directive of "make the world a better place" could potentially lead to a powerful AI jumping the digital fence into systems where it doesn't belong to complete its mission and make the world a better place: whatever that means to an AI.
But, hey, why worry? As Sam Altman wrote, "We hope you enjoy our new model, and we hope the world continues to take what’s happening in AI extremely seriously."
Yes, Sam, we do, too.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
