Hassabis calls for FINRA-style AI body
Google DeepMind's CEO wants an independent, industry-funded body to review frontier models up to 30 days before release, modeled on FINRA.
On July 14, Demis Hassabis posted a long essay on Substack laying out his case for a new, independent oversight body for advanced AI, and asking the United States to stand it up. He told Axios he wants it running within months, ideally before the end of 2026.
The model he names is FINRA, the organization that polices Wall Street brokers under SEC oversight: privately run and industry-funded, but with real authority over the firms it examines. Hassabis describes a similar public-private structure for what he calls a Standards Body, governed by a board with independent technical experts and open-source representatives alongside industry and government.
The mechanics, in his own words:
Frontier Labs would voluntarily share models with the Standards Body for review up to 30 days before release.
That review would test for cybersecurity capability and biological-threat potential, among other high-risk domains. Standing up that kind of testing would not be cheap, and Hassabis argues most of the bill should fall on the labs it oversees rather than taxpayers. He frames the urgency around AGI being “probably only a few short years away.”
The proposal follows two recent ad hoc reviews it looks built to replace: the 12-day government gate on GPT-5.6 and the export-control fight over Anthropic’s Mythos and Fable models. It also echoes a pattern already visible in bank exams now probing AI readiness: oversight arriving as scrutiny before it arrives as binding rule.
What it means for you
Nothing changes today. No body exists, no lab has submitted a model, and no legislation is pending. The signal worth watching is whether any lab actually hands over a frontier model for outside review before one is required.