Today’s AI-driven cyber risks were “warning shots,” Hassabis told us from his London base. Within 18 months, he said, those capabilities — plus far more serious biological and nuclear threats — could be enshrined in open-source models beyond any government’s control.
- Hassabis emphasized to us that risks will come from the larger labs’ more powerful future proprietary models, not just open source models.
- “What we do together now,” he writes in his manifesto, “will determine how the next phase of civilization unfolds.”
Behind the scenes: Hassabis spent months quietly building support for the plan, briefing the Trump administration, other lab leaders and European officials before it went public.
- “The sounds I’ve heard are very positive,” he said of his discussions with the government, which had taken a laissez-faire approach to AI regulation before the myth scare.
- Hassabis, a scientist who enjoys rare respect in the warring AI camps, says the other major lab leaders agree at a high level: “This is where the industry needs to go.”
- His schedule is aggressive. “Months,” said Hassabis, ideally if the new body is up and running “before the end of the year.”
This is how it works: Hassabis proposes an AI standards board modeled after FINRA (Financial Industry Regulatory Authority), the private, industry-funded regulator that oversees Wall Street under the supervision of the SEC.
- Frontier Labs would initially voluntarily share its models with the body up to 30 days before release to conduct security tests examining dangerous cyber, biological and “deception” capabilities.
- According to Hassabis, once the testing system proves to be “effective and robust,” “formalization could occur quickly.” This means that borderline models must be approved before they can be used in the US market.
- Hassabis envisions a majority independent board made up of Turing Award winners and other recognized experts, as well as representatives from industry, government and open source.
The intrigue: The rules would apply to all frontier models “regardless of their country of origin or whether they are open or closed” – with qualifying benchmarks updated regularly as capabilities evolve.
- Hassabis predicts that the term “borderline” would carry a seal of approval: being tested means you are important. “I think that’s a pretty nice, prestigious benefit,” he told us.
The big picture: The Trump administration’s improvised crackdown on Anthropic’s Mythos and Fable models last month was “kind of a wake-up call,” Hassabis said – evidence that Washington needs something more robust than ad hoc policies.
- Anthropic saw its best-performing models frozen overnight by an export control order and then spent two and a half weeks negotiating their release with no set rules, protocols or rules of the game.
- Hoping to avoid the same fate, OpenAI agreed to limit GPT-5.6 to government-certified partners at launch. It was released publicly last week after negotiations and testing with the Commerce Department.
Between the lines: Anthropic CEO Dario Amodei has himself called for mandatory regulation, envisioning an FAA-style agency with the power to block unsafe models.
- The lab leaders behind Gemini and Claude now agree that Washington should regulate them, although they mainly disagree about who has the authority.
The conclusion: Hassabis believes that AGI – a system with all the cognitive capabilities of the human brain – is “probably only a few years away” and that we are “at the foot of the singularity.”
- “We essentially found a way to make sand think,” he writes. “It’s a miracle.”
Read the post.
https://www.axios.com/2026/07/14/demis-hassabis-ai-regulation-google-deepmind
