DeepMind proposes a standard for releasing frontier AI

Google DeepMind CEO Demis Hassabis has proposed creating an independent body to assess frontier models before release.

The idea follows a FINRA-style model: voluntary reviews 30 days before launch, which could later become mandatory for U.S. market access if the approach proves effective.

The system would include technical experts, representatives from open source and industry, and could be funded by the AI sector itself.

Hassabis says this could help balance innovation and accountability, while enabling faster responses to critical post-release risks.

Why it matters: AI is advancing faster than regulation can adapt, and strong oversight could reduce risk without slowing innovation unnecessarily.

Do you think AI needs a dedicated supervisory body? #AI #DeepMind #regulation