Demis Hassabis proposes U.S. AI watchdog for frontier models
AI safety testing body could review frontier models before public release Get clearer risk checks on deception, cyber abuse, and misuse as AI advances
Google DeepMind CEO Demis Hassabis has proposed a U.S.led oversight body that would safetytest advanced AI models before they are released to the public. His plan would focus on frontier systems and aims to establish clearer rules for reviewing risks such as deception, cyber abuse, and bioweaponrelated misuse.
According to the proposal, companies developing the most capable models would voluntarily submit them for review about 30 days before launch. Hassabis said the approach should be adaptable as AI capabilities change and suggested it could coordinate a slowdown among frontier labs if regulators judged that necessary.
The idea comes as governments and companies continue to debate how AI should be governed. Hassabis wants the body operating this year, arguing that opensource systems could reach more dangerous capabilities within the next 18 months.