Demis Hassabis, co-founder and CEO of Google DeepMind, has called for the United States to establish a new global AI oversight body with authority to screen the world’s most advanced models and to coordinate an industry-wide slowdown if risks escalate. He set out the proposal in a personal manifesto titled "A Framework for Frontier AI and the Dawning of a New Age," published Tuesday morning.
Why this matters
In an exclusive interview with Axios, Hassabis argued for a more systematic regulatory approach: an industry-funded organization staffed with world-class technical experts and accountable to the U.S. government. He described current AI-driven cyber threats as "warning shots," and warned that within 18 months such capabilities — along with graver biological and nuclear threats — could appear in open-source models beyond the control of any government.
Hassabis also emphasized that significant risks are likely to come from the major labs' more powerful future proprietary models, not only from open-source systems. In his manifesto he wrote, "What we collectively do now will determine how the next phase of civilization unfolds."
Political and industry context
Hassabis spent months quietly building support for the idea, briefing the Trump administration, fellow lab leaders and European officials before going public. He described his talks with the administration as "very positive," noting that Washington’s prior approach to AI regulation had been relatively laissez-faire before the Mythos episode.
The immediate catalyst cited is the improvised U.S. response to Anthropic’s Mythos and Fable models: an export-control order that froze Anthropic’s most powerful models overnight, followed by 2½ weeks of negotiations conducted without established rules, protocol or a playbook. Hassabis said that episode was a wake-up call demonstrating the need for something more durable than ad hoc directives.
OpenAI, seeking to avoid a similar outcome, initially restricted GPT-5.6 to government-vetted partners at launch and only released it publicly last week after testing and negotiations with the Commerce Department.
How the proposed watchdog would work
Hassabis proposes a standards body modeled on FINRA (Financial Industry Regulatory Authority): an industry-funded watchdog operating under U.S. oversight. Key elements of his plan include:
- Frontier labs would voluntarily share their models with the body up to 30 days before release for safety testing focused on dangerous cyber, biological and "deception" capabilities.
- Once the testing regime proves effective and robust, formalization could follow quickly, meaning frontier models would be required to pass tests before being deployed in the U.S. market.
- The oversight board would be majority-independent and include Turing Award winners and other credentialed experts alongside industry, government and open-source representatives.
The rules would apply to all frontier-class models regardless of country of origin or whether they are open or closed, and qualifying benchmarks would be updated as capabilities evolve. Hassabis predicted that being labeled and tested as a "frontier" model would confer prestige: "I think that's a pretty nice, prestige kind of asset to have," he told Axios.
Timing and reception among lab leaders
Hassabis set an aggressive timeline: "months," and ideally the new body would be operational "before year-end." He said other major lab leaders broadly agree on the industry’s need to move in this direction, though they differ on who should hold final authority.
Anthropic CEO Dario Amodei has separately called for binding regulation and envisions an FAA-style agency with power to block unsafe models.
The big picture
Hassabis warned that artificial general intelligence (AGI) — a system with the full suite of human cognitive powers — is "probably only a few short years away," and said humanity stands in "the foothills of the singularity." He described the technological advance with a colorful phrase in his manifesto: "We've essentially found a way to make sand think. It's miraculous."
Further details and rationale are laid out in Hassabis’s published manifesto.
