DeepMind's Demis Hassabis calls for a US-led global AI watchdog

Demis Hassabis used a new essay to argue that frontier AI governance needs a standards body modeled loosely on financial self-regulation. The proposed US-led organization would bring independent experts and open-source representatives into model evaluation, with enough authority to pause or slow releases that present unacceptable risks. The push comes as Hassabis says AGI may be only a few years away and as governments are taking a more active role in model-release decisions.

Key details: Hassabis proposed a US-led standards body for frontier AI model reviews, The body would be able to coordinate an industry slowdown for models judged too risky, The proposal follows recent government scrutiny of advanced model releases.

Why it matters: One of the leading frontier-lab CEOs is moving from general safety warnings to a concrete release-review institution with enforcement teeth.