Updated
Updated · TechCrunch · Jul 14
Hassabis Urges 30-Day AI Model Reviews by Independent U.S.-Backed Standards Body
Updated
Updated · TechCrunch · Jul 14

Hassabis Urges 30-Day AI Model Reviews by Independent U.S.-Backed Standards Body

3 articles · Updated · TechCrunch · Jul 14

Summary

  • Demis Hassabis proposed that frontier AI labs voluntarily submit models for review up to 30 days before release, with deployment in the U.S. later tied to passing the assessment.
  • The Google DeepMind chief said the body should be modeled on FINRA—government-backed, industry-funded and independently run—to test models, set release practices and handle post-release vulnerabilities.
  • His plan responds to criticism of recent U.S. government reviews of Anthropic's Mythos and OpenAI's Sol, which were faulted for limited technical expertise and opaque release decisions.
  • The proposal also tries to navigate Trump administration resistance to a formal in-house regulator after White House AI adviser Sriram Krishnan said there would be no 'FDA for AI.'
  • Hassabis said the body could include open-source representatives, industry experts and outside safety groups, creating a flexible framework that can tighten as frontier-model risks grow.