Google DeepMind CEO Proposes US Standards Body to Test AI Models Before Release

Demis Hassabis, CEO of Google DeepMind, has proposed the establishment of a US standards body that would vet the most powerful AI models prior to their public release. The proposed evaluations would focus on cybersecurity, biological threats, and attempts by systems to bypass restrictions, mislead users, or conceal their actions.

Initially, companies would voluntarily submit models for assessment no later than 30 days before launch. If the mechanism proves effective, Hassabis suggests making the review mandatory for market access in the United States.