Quick Facts
- Google DeepMind CEO Demis Hassabis proposed a federally supervised AI standards body modeled on FINRA, the private watchdog that polices Wall Street under SEC oversight.
- Frontier labs would voluntarily share models up to 30 days before release for review, with mandatory compliance once the testing regime proves effective.
- Hassabis has held talks about the initiative with the White House, European officials, and competing AI labs, with a target launch before year-end.
Google DeepMind CEO Demis Hassabis on Tuesday called for a U.S.-led standards body to test new frontier AI models for national security risks before they reach the market. Hassabis laid out the plan in a personal manifesto published Tuesday on Substack, titled “A Framework for Frontier AI and the Dawning of a New Age.”
The proposed organization would function as a federally supervised public-private partnership, or alternatively as a self-regulatory body modeled after FINRA, the nonprofit watchdog funded by the companies it regulates and supervised by the U.S. Securities and Exchange Commission. Its board would include a majority of independent technical experts, with Turing Award winners, open-source sector representatives, and government officials also seated.
Hassabis wants the body running fast. “Months,” he said, with the organization operational before year-end.
How It Would Work
Under the proposal, frontier labs would initially share models with the body for review up to 30 days before public release on a voluntary basis. Once the testing regime proves effective, passing that review would become mandatory for any deployment in the U.S. market.
The rules would apply to all frontier-class models, regardless of country of origin or whether they are open or closed source. Qualifying benchmarks would be updated quarterly, with the cadence able to accelerate as capabilities advance. Funding would likely come from industry, though Hassabis said the body would need “substantial” resources to attract top technical talent and provide compute for large-scale testing.
Hassabis argues a prestige incentive would drive compliance. “I think that’s a pretty nice, prestige kind of asset to have,” he told Axios, suggesting that a frontier designation from the body would carry market weight.
What Prompted the Push
The Trump administration’s improvised crackdown on Anthropic’s Mythos and Fable models last month served as a catalyst. Hassabis called the episode “a bit of a wake-up call.” Anthropic saw its most powerful models frozen overnight by an export-control order, then spent two and a half weeks negotiating their release with no established rules or protocol.
OpenAI, hoping to avoid the same outcome, agreed to restrict GPT-5.6 to government-vetted partners at launch before public release last week following negotiations with the Commerce Department. Both episodes drew criticism for opaque decision-making and a lack of technical expertise in the review process.
The proposal also arrives as Chinese AI models from companies including DeepSeek and Z.ai gain traction among U.S. businesses. State Department officials told CNBC that growing adoption of Chinese AI models raises “serious concerns.” Hassabis said that within 18 months, advanced AI capabilities, including biological and nuclear threat vectors, could live inside open-source models beyond any government’s control. Today’s AI-driven cyber risks, he told Axios, are “warning shots.”
Industry Alignment
Hassabis is not alone among lab chiefs calling for regulation. Anthropic CEO Dario Amodei has called for binding regulation modeled on the FAA, with authority to block unsafe models outright. The two executives differ mainly on who holds the power, not whether oversight should exist.
Hassabis framed the urgency in stark terms in his manifesto. “AGI cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile,” he wrote. “It is much more akin to the discovery of electricity or fire.” He added: “What we collectively do now will determine how the next phase of civilization unfolds.”
Read more: Google DeepMind CEO Demis Hassabis calls for creation of AI standards body