Demis Hassabis has proposed creating a global artificial intelligence watchdog led by the United States to regulate frontier models before they are released. Writing in a blog post, the Google DeepMind chief executive argued that the US should direct this initiative due to its economic and technical standing. He described the body as an organisation of independent experts and open-source representatives with the power to halt deployment if risks become too high. This structure would function similarly to existing financial regulators but would apply specifically to advanced AI systems. Hassabis believes current national approaches are insufficient to manage the scale of potential harm from these technologies.
The suggestion highlights a growing consensus that international coordination is necessary despite the lack of a unified framework. It places the onus on the country currently dominating model development to set the rules for the entire industry. The proposal also touches on the tension between innovation speed and safety verification. Hassabis emphasises that early intervention is preferable to reacting after systems cause damage.
* The watchdog would evaluate models prior to public release.
* Membership would include independent experts and open-source community representatives.
* The US is selected to lead due to its current technical dominance.
