DeepMind CEO Hassabis proposes FINRA-style standards body for frontier AI

An article reports that Google DeepMind chief executive Demis Hassabis used an X post on a Tuesday morning to urge the creation of a new body to oversee the release of frontier AI models. The post describes a "standards body" patterned on the Financial Industry Regulatory Authority (FINRA) that would test frontier models and develop release best practices.

According to the article, the proposal would draw on informal US government reviews of two models, Mythos (Anthropic) and Sol (OpenAI). Critics faulted those reviews for a shortage of technical expertise and for unclear decisions over release timing. Under Hassabis' idea, such decisions would move to a new organization that the US government backs, the AI industry pays for, and that operates independently.

Participation would begin as voluntary: labs would submit models to the body for review as much as 30 days ahead of release. If the assessment protocol proved effective and robust, it could be formalized quickly, so that frontier models would have to pass it before being deployed in the US market. The body would also work with labs on serious post-release vulnerabilities. Hassabis imagines staff drawn from open source representatives and industry technical experts, financed by the labs, with some evaluations outsourced to specialized AI safety groups.

The article also notes the politics: White House AI advisor Sriram Krishnan recently ruled out an executive-branch regulator, declaring that no "FDA for AI" is coming. It presents the self-regulatory structure as one way to answer such objections. Hassabis argues the approach would be technically focused while supporting innovation and rewarding responsible behavior, and could be tightened if needed.