Google DeepMind Founder Proposes AI Watchdog

Demis Hassabis, co-founder of Google DeepMind, has proposed the creation of a new U.S.-led oversight body to address safety concerns surrounding advanced AI models.

Hassabis suggests this new organization, which would be industry-funded and federally overseen, could begin operating by the end of 2026. Its mandate would be to test the world's most sophisticated AI models before their release, potentially barring unsafe systems from the American market.

The proposal, outlined in a Substack post, aims to replace ad hoc government interventions with clearer, standardized rules. Hassabis drew a parallel to the Financial Industry Regulatory Authority (FINRA), Wall Street's self-regulatory body.

Initially, AI developers could voluntarily submit models for review up to 30 days prior to launch. These tests would assess the models' potential to facilitate cyberattacks, biological threats, or deceptive behavior. Future iterations may make review mandatory for certain frontier models, including foreign and open-source systems entering the U.S. market.

"We believe the standards body can begin as an industry-led, voluntary organization, that in future gets formalized by statute or executive order," stated Owen Larter, Google's senior director of frontier AI policy and global affairs.