Demis Hassabis, CEO of Google DeepMind, has called for the United States to lead the creation of a global artificial intelligence watchdog. This organization would possess the authority to screen advanced AI models and coordinate industry-wide slowdowns if potential dangers emerge.
Google DeepMind CEO Demis Hassabis has proposed the establishment of a United States-led global body to regulate advanced artificial intelligence. This agency would be tasked with evaluating frontier AI models and could halt development if significant risks are identified. Hassabis, in a personal manifesto published on Tuesday, outlined a vision for a body modeled after financial regulators like the Financial Industry Regulatory Authority (FINRA). Such an organization, he argued, would be funded by the AI industry itself, staffed by leading technical experts, and accountable to the U.S. government.
Hassabis expressed concern that the rapid pace of AI development, particularly with frontier models, could lead to unforeseen and potentially catastrophic consequences. He warned that within 18 months, dangerous capabilities, including those related to biological and nuclear threats, could emerge in unregulated open-source models. Hassabis has previously referred to the current period as a "species-level transition" for humanity, suggesting that AI advancements are occurring at an unprecedented speed, potentially outpacing humanity's ability to manage them. He believes that artificial general intelligence (AGI), or AI that matches human cognitive abilities, is likely only a few years away.
The proposed watchdog would conduct pre-release testing of AI models, requiring companies to demonstrate safety before deployment. This testing would initially be voluntary for "Frontier Labs" in the U.S. but could eventually become mandatory, with the aim of extending the framework globally. The evaluations would focus on identifying dangerous cyber, biological, and deception capabilities. Hassabis suggested that if necessary, the body could coordinate a slowdown in development, drawing a parallel to recent considerations by companies like Anthropic. He stressed that the regulations should apply to all frontier-class models, irrespective of their country of origin or whether they are open-source or proprietary.
Hassabis has reportedly discussed his proposal with Trump administration officials and European leaders, who have responded positively to the idea. He believes that other leading AI labs are in general agreement with his proposal, stating, "This is where the industry needs to go".
However, the proposal also raises questions about its potential impact on smaller companies and open-source development. Critics suggest that mandatory pre-release testing and evaluation could disproportionately burden startups and smaller competitors, while potentially constraining innovation in the open-source AI community. Hassabis, however, has indicated that non-frontier models from startups or academic research would be exempt from these requirements.
Hassabis's call for regulation comes amid growing debate about AI safety and the potential risks associated with rapidly advancing AI technologies. He has previously warned about "race conditions" in AI development, where competitive pressures might lead labs to release models before fully verifying their safety. The proposal also follows recent actions by the Trump administration concerning AI models, which Hassabis described as a "wake-up call".