DeepMind CEO Demis Hassabis proposed creating an independent international body
to test and review frontier AI models before public release and will meet US
officials in Washington next week to discuss the plan. Hassabis cited advanced
cyber-attack capabilities demonstrated by ANTHROPIC's MYTHOS as evidence of the
need for stronger AI safety oversight. The proposal would place reviews in the
hands of independent experts, be funded by major AI firms, allow up to 30 days
of testing per model, and include provisions to coordinate industry slowdowns of
frontier model development if necessary. OpenAI CEO Sam Altman and Elon Musk
have publicly expressed support.
