DeepMind's Hassabis Urges U.S. AI Standards Body With 30-Day Model Reviews

3 articles · Updated · CNBC · Jul 14

Demis Hassabis said the U.S. should create a federally overseen AI standards body to review frontier models before release and assess cybersecurity, bio and other national-security risks.
His proposal would start with voluntary submissions up to 30 days before launch, then make reviews mandatory for models deployed in the U.S. once the system proves effective.
Hassabis said the body should resemble FINRA, include independent technical and open-source representatives, and receive substantial industry-backed funding for talent and large-scale testing.
The push comes as Washington and AI firms clash over oversight—Anthropic and OpenAI recently faced U.S. restrictions—and as lawmakers weigh curbs on Chinese models gaining traction in the U.S.