Demis Hassabis Proposes Wall Street-Style Referee for AI
Google DeepMind CEO calls for global oversight in critical pre-AGI period.
1Global AI Oversight System
Google DeepMind CEO Demis Hassabis proposed a Wall Street-style regulator to vet frontier AI models before release. Inspired by FINRA, the system would operate with a board of independent experts to evaluate advanced AI systems.
2From Voluntary to Mandatory Oversight
Initially, labs would voluntarily submit models 30 days pre-release for testing cyberattack capabilities, biological/nuclear risks, and deception signs. The regime would gradually become mandatory as the system matures.
3Emergency Brake Mechanism
The most striking proposal is a ratcheting mechanism to slow development pace as risks escalate. Hassabis suggests enabling 'coordinated slowdown among Frontier Labs when necessary' - an unusual suggestion from a lab CEO.
418-Month Critical Warning
Hassabis characterizes current cyber risks as 'warning shots,' warning that far more dangerous biological and nuclear tools could emerge in open-source models within 18 months - risks governments couldn't recall.
5Growing Industry Demand for Regulation
Anthropic's export ban and OpenAI's GPT-5.6 holdback amplified regulatory demands. Hassabis and Anthropic's Dario Amodei previously called for a US-led coalition at G7, though they differ on regulatory approaches.
6The Age of Thinking Sand
Hassabis frames AGI as fundamental as fire or electricity: 'We've essentially found a way to make sand think - it's miraculous.' Yet questions remain whether labs-funded oversight can impartially halt profitable models.