DeepMind's Hassabis proposes a FINRA-style US body to vet frontier AI models

Google DeepMind chief executive Demis Hassabis published an essay proposing a new US standards body to review frontier AI models before release, modelled on FINRA, the industry-funded regulator that polices Wall Street brokers under SEC oversight. Under the proposal, “Frontier Labs” — companies whose models cross defined capability thresholds — would initially share models with the body voluntarily, up to 30 days before release, for evaluation of cyber, biological and deception-related risks; participation could later become a condition of operating in the US market.

Hassabis argued the case for urgency rather than caution: he described artificial general intelligence as “probably only a few short years away” and its likely impact as comparable to “the discovery of electricity or fire,” framing competitive pressure between labs and between nations as outpacing the field’s ability to understand its own technology. His essay proposed quarterly updates to the benchmarks used to define “frontier” status, and suggested the body could draw technical staff from AI companies, open-source developers and specialist evaluation groups such as those already used for ad hoc government reviews of models including Anthropic’s Mythos and OpenAI’s Sol — reviews that had drawn criticism for lacking technical depth and transparency.

Reaction from rival executives was notably positive for an industry usually defined by competition: OpenAI’s Sam Altman called the proposal “thoughtful,” Elon Musk described it as “a thoughtful framework overall and certainly a good starting point for discussions,” Microsoft’s Satya Nadella offered public praise, and Anthropic co-founder Jack Clark called it “excellent.” The White House was more guarded — an AI adviser had separately said there would be no “FDA for AI” — though press reporting said the Trump administration was independently weighing a similar self-regulatory structure.

The episode marked one of the few moments where frontier-lab leaders converged publicly on a specific governance mechanism, after two years in which US federal legislation on frontier models had stalled. Commentators noted that public praise for a voluntary, industry-shaped body was not the same as agreement to binding, government-run oversight, and that the proposal’s substance — who would staff it, and what would trigger mandatory compliance — remained unresolved.