Google AI chief calls for US-led global safety watchdog
Demis Hassabis said a global AI oversight body should review frontier models and have power to delay releases deemed too risky.
By James Whitfield · Staff Writer
2 min read
Google DeepMind chief executive Demis Hassabis has called for a global AI oversight body with authority to slow or block the release of powerful new models if they appear too risky. In a LinkedIn post, Hassabis said the United States should lead the effort because of its economic and technical position.
The proposal adds a prominent industry voice to calls for stronger oversight of frontier AI systems, the most advanced models being built by major labs. Hassabis said the body should help set global standards and examine frontier models before they are released.
What Hassabis proposed
Hassabis, who co-founded Google DeepMind, said the organization could be modeled on existing regulators such as the Financial Industry Regulatory Authority. He described a body made up of independent experts and representatives from open-source communities.
Under the framework he outlined, the group would have the ability to evaluate advanced AI models ahead of release. Hassabis also said it should have power to intervene if a model reaches a level of danger that warrants stopping or delaying deployment.
He wrote that the US should take the lead in creating the organization, saying the country is well placed to set international norms “given its economic and technical standing.” The Verge reported the remarks from Hassabis’s LinkedIn post.
The reference to open-source communities signals that Hassabis is not calling for a body limited to large corporate AI labs. His proposal would include people connected to open model development alongside independent specialists.
Why the proposal matters
Hassabis runs one of the companies building frontier AI models, giving the proposal weight inside the industry. Google DeepMind is part of Google’s broader AI push and competes with other labs developing increasingly capable systems.
The idea would place outside review ahead of some model launches, rather than relying only on internal company testing. Hassabis framed that authority as a way to respond if future systems show signs of becoming unsafe.
The post did not provide a full design for how such a global watchdog would be created, how its decisions would be enforced across borders or which governments would participate. Hassabis’s central point was that international AI oversight should begin with US leadership and include expert review before frontier systems reach the public.
This story draws on original reporting from The Verge.