Demis Hassabis, co-founder and CEO of Google DeepMind, has published a manifesto calling for the United States to establish a dedicated watchdog to evaluate frontier artificial intelligence models before they are deployed. The proposed body, modeled after the financial industry’s FINRA, would be privately funded by AI labs, staffed with independent technical experts, and answerable to the government. Hassabis wants the agency operational before the end of the year, warning that artificial general intelligence (AGI) — which he defines as a system matching the full range of human cognitive ability — is “probably only a few short years away,” with a timeline around 2030.
Under his plan, frontier labs would voluntarily hand over their models up to 30 days before release so the body could test for dangerous cyber, biological, and “deception” capabilities. Once testing proves reliable, formal rules “could quickly follow,” making pre‑deployment safety clearance mandatory for any frontier‑class system entering the U.S. market, regardless of its country of origin or whether it is open‑source. The board would be majority‑independent, filled with Turing Award winners and other credentialed researchers, alongside industry, government, and open‑source developer representatives. Hassabis argued that the “frontier” designation would become a prestigious mark of status.
The urgency stems from what Hassabis describes as “warning shots”: today’s AI‑driven cyberattacks that presage more serious biological and nuclear risks within 18 months. He noted that the danger is not limited to open‑source models — proprietary systems will carry the same threats. In recent weeks, he has briefed the Trump administration, rival lab leaders, and European officials, saying feedback from the White House has been “very positive.” The administration recently placed Anthropic’s Mythos and Fable models under an export‑control order (later lifted with extra controls), while OpenAI voluntarily limited GPT‑5.6 to vetted partners before releasing it after testing with the Commerce Department. Hassabis called the incident “a bit of a wake‑up call.” Anthropic CEO Dario Amodei also supports a regulatory body but favors an FAA‑style agency; Hassabis prefers the lighter FINRA structure.
In the manifesto, Hassabis wrote that researchers have “found a way to make sand think,” calling it “miraculous.” He stressed that the future is not yet written and urged action in the “precious window before AGI arrives.”
