Hassabis Urges US-Led AI Watchdog to Screen Frontier Models

Advanced AI systems now carry concrete risks of enabling cyberattacks, biological weapons and nuclear threats that could escape any single government's reach within 18 months. Demis Hassabis, Google DeepMind CEO and 2024 Nobel laureate in Chemistry, argues that only a new US-led standards body can impose systematic safety checks before those capabilities spread.

Hassabis laid out the proposal in a manifesto titled "A Framework for Frontier AI and the Dawning of a New Age," published July 14, 2026. The body would operate like FINRA, funded by industry yet overseen by the US government, with a majority-independent board of Turing Award winners and technical experts. Frontier labs would submit models for testing of cyber, biological and deception risks up to 30 days before release; once the regime proves effective, passing the tests would become a requirement for US market access. Benchmarks defining "frontier" status would update as capabilities advance, applying equally to open and closed models regardless of origin.

Hassabis told Axios the plan has received positive signals from the Trump administration after months of private briefings that also reached other lab leaders and European officials. He set an aggressive timetable: the organization should be operational before year-end. Current ad hoc export-control actions, such as the overnight freeze on Anthropic's Mythos and Fable models and OpenAI's subsequent restrictions on GPT-5.6, illustrate the absence of established rules, he said. Hassabis described today's AI-driven cyber incidents as "warning shots" and placed AGI, systems matching or exceeding human cognitive abilities, only a few years away.

Anthropic CEO Dario Amodei has separately advocated an FAA-style agency with blocking authority. No global AI treaty or comprehensive US statute yet exists. Hassabis signed a statement last month urging stronger safeguards against AI-aided bioweapons. The proposal's details on enforcement mechanics and international coordination remain at the conceptual stage.