ทันเอไอ

Google DeepMind CEO proposes US-led 'AI Regulatory Body' to set global safety standards

Demis Hassabis, CEO and co-founder of Google DeepMind, has called for the US to establish a frontier AI oversight agency with the authority to audit and pause development if dangerous risks are detected.

Google DeepMind CEO Demis Hassabis has proposed the creation of a US-led global regulatory body to audit and oversee 'frontier' AI models. If these cutting-edge systems pose excessive risks, the agency would have the power to halt their release. The proposal was detailed in his July 14th essay, 'A Framework for Frontier AI and the Dawning of a New Age.'

Hassabis suggests modeling this new agency after FINRA, the self-regulatory organization for the US financial industry. Initially, leading AI labs would voluntarily submit their models for a 30-day safety review—testing for cybersecurity, biological threats, and 'deception' capabilities—prior to public release. Once the framework is validated, all frontier models would be required to pass this assessment before entering the US market, regardless of origin or whether they are open or closed-source. He aims for the agency to be operational by the end of this year.

The push for this framework follows a recent incident where the Trump administration suddenly moved to restrict the export of Anthropic’s Mythos and Fable models. Hassabis views this as a 'warning sign' of the need for clear, consistent mechanisms over ad-hoc government interventions, noting that Anthropic spent over two weeks in negotiations without a clear regulatory roadmap.

Hassabis is not alone in this call. Anthropic CEO Dario Amodei has similarly advocated for binding regulations, drawing parallels to the FAA’s authority to ground unsafe aircraft. This reflects a growing consensus among top AI leaders that standardized oversight is essential to ensuring safety in an era of increasingly powerful AI systems.