Hassabis Calls for U.S.-Led AI Watchdog to Screen Frontier Models

Google DeepMind co-founder and CEO Demis Hassabis is calling on the United States to establish a new AI watchdog with the power to screen the world's most advanced models — and coordinate an industry-wide slowdown if dangers mount ¹.

Hassabis, a Nobel laureate, laid out the proposal in a personal manifesto published Tuesday morning titled "A Framework for Frontier AI and the Dawning of a New Age" ¹. He also detailed the plan in a Substack essay proposing that the U.S. should lead the effort ².

The proposed body would be modeled on FINRA, the financial industry's self-regulatory organization ¹. Under the plan, frontier labs would initially share their models voluntarily with the body up to 30 days before release for safety testing ¹. The safety testing would probe dangerous cyber, biological, and "deception" capabilities ¹. Frontier models would be required to pass before they could be deployed in the U.S. market ¹. The qualifying benchmarks would be regularly updated as capabilities evolve ¹.

The rules would apply to all frontier-class models regardless of country of origin or whether they are open or closed ¹. Hassabis envisions a majority-independent board comprising Turing Award winners, credentialed experts, and representatives from industry, government, and the open-source community ¹. He said the frontier designation would carry cachet and serve as a prestige asset ¹.

Hassabis told Axios that the time has come for a more "systematic" approach to AI regulation — funded by the industry, staffed by world-class technical experts, and answerable to the U.S. government ¹. He said he wants the new body operational before year-end ¹.

In the interview, Hassabis characterized today's AI-driven cyber risks as "warning shots" ¹. Within 18 months, he said, those capabilities — plus far graver biological and nuclear threats — could live inside open-source models beyond any government's control ¹. He emphasized that risks will come from major labs' more powerful future proprietary models, not just open-source models ¹.

"What we collectively do now will determine how the next phase of civilization unfolds," Hassabis wrote in the manifesto ¹. He said he believes AGI is probably only a few short years away and described the current moment as standing in "the foothills of the singularity" ¹.

Hassabis has spent months quietly building support for the plan, briefing the Trump administration, fellow lab leaders, and European officials before going public ¹². "The noises I've been hearing are very positive," he said of his talks with the administration ¹. He said other major lab leaders agree at a high level that this is where the industry needs to go ¹.

The proposal arrives amid recent regulatory turbulence. The Trump administration's crackdown on Anthropic's Mythos and Fable models last month was described by Hassabis as a wake-up call ¹. Anthropic's most powerful models were frozen overnight by an export-control order, and the company spent two and a half weeks negotiating their release with no established rules, protocol, or playbook ¹. GPT-5.6 was released publicly last week after negotiations and testing with the Commerce Department, with OpenAI agreeing to restrict the model to government-vetted partners at launch ¹.

Anthropic CEO Dario Amodei has called for binding regulation and an FAA-style agency with power to block unsafe models ¹. [ANALYSIS] The lab chiefs behind Gemini and Claude now converge on the principle that Washington should regulate frontier AI, with the central disagreement centering on institutional design — a self-regulatory body modeled on FINRA versus a government agency modeled on the FAA. Hassabis said once the testing regime proves effective and robust, formalization could quickly follow ¹.