Hassabis urges U.S.-led AI watchdog to slow dangerous frontier models

Demis Hassabis called for a U.S.-led AI watchdog with the power to screen the world’s most advanced AI systems and, if needed, coordinate an industry-wide slowdown. In a blog post published on July 14, 2026, the Google DeepMind chief executive framed the idea as a response to frontier models that could create severe cyber and biological risks.

Hassabis argued that the United States was best placed to lead because of its economic position and its role in shaping the market for advanced systems.

The watchdog Hassabis described would evaluate frontier models before deployment, draw on independent experts and representatives from open-source communities, and step in if model risks escalated.

Hassabis has been sharpening the warning around that push for months. He has said humanity is in a “precious window” to make AGI safe and has argued that advanced AI may be only a few years away. In earlier public remarks, he described AI as a “species-level transition” with “little margin for error,” language that now sits behind his call for faster governance.

On June 17, 2026, Hassabis and Anthropic chief executive Dario Amodei called for a U.S.-led AI coalition during a closed-door meeting at the G7 summit in Évian-les-Bains, France. President Donald Trump attended that lunch, where tech leaders and heads of state discussed coordination on AI rules and standards.

The company published the third update to its Frontier Safety Framework in 2025 to identify and mitigate severe risks from advanced AI models. OpenAI proposed frontier AI regulation that would require pre-deployment risk assessments, external scrutiny of model behavior and the use of those assessments to guide deployment decisions.

OpenAI, Google, Anthropic and Microsoft also formed the Frontier Model Forum in 2023 to advance frontier AI safety research and standards.