Google DeepMind CEO Demis Hassabis proposes US-led global AI watchdog

The chief executive of Google DeepMind, Demis Hassabis, has formally proposed the establishment of a United States-led global watchdog designed to police the development of advanced artificial intelligence. Writing in a manifesto titled "A Framework for Frontier AI and the Dawning of a New Age," Hassabis argued that the current regulatory vacuum necessitates a more systematic approach to managing the risks posed by "frontier" models.

Hassabis envisions a regulatory body modeled after the Financial Industry Regulatory Authority (FINRA). Under this structure, the watchdog would be industry-funded and staffed by independent technical experts, yet remain answerable to the U.S. Government. The proposed agency would possess the authority to conduct safety screenings on the most powerful AI systems before they are released to the public and coordinate industry-wide slowdowns if risks are deemed too severe.

Media additions

Addressing the "Precious Window"

The push for centralized oversight comes amid fears that the arrival of artificial general intelligence (AGI) is imminent. Hassabis, a 2024 Nobel Prize winner, stated that humanity is standing in the "foothills of the singularity" and that AGI could arrive within a few years. He emphasized that the current period represents a "precious window" to implement safeguards before capabilities exceed the capacity of existing government oversight. He further described current AI-driven cybersecurity risks as "warning shots," noting that within 18 months, biological and nuclear threats could reside within open-source models if left unregulated.

The proposal arrives following recent friction between technology firms and Washington. Last month, the Trump administration intervened in the releases of models from Anthropic—specifically the Mythos and Fable systems—which resulted in abrupt, ad-hoc export control freezes. Hassabis described these events as a "wake-up call" indicating that policymakers require a more robust, predictable framework than reactive directives. He noted that Anthropic spent over two weeks negotiating the release of its models without established protocols.

Proposed Operational Framework

According to reports, the proposed regulatory body would operate through the following mechanisms:

• Voluntary Submission: Labs would initially submit frontier models for safety testing at least 30 days prior to public release.
• Broad Jurisdiction: Rules would apply to all frontier-class models, regardless of whether they are open or closed, or their country of origin.
• Testing Focus: Evaluations would prioritize cyber, biological, and "deception" capabilities.
• Formalization: Once proven effective, the voluntary regime could transition into a mandatory requirement for models entering the U.S. Market.

Industry and Political Reactions

Hassabis has spent several months briefing the Trump administration, rival lab leaders, and international officials on the plan. He characterized the feedback from the Trump administration as "very positive," despite its prior preference for a laissez-faire approach to technology governance. Other leaders, including Anthropic CEO Dario Amodei, have also expressed support for binding, systematic regulation, with Amodei suggesting an FAA-style agency. At a G7 meeting held in Évian-les-Bains, France, Hassabis and Amodei reportedly backed a U.S.-led coalition alongside OpenAI chief executive Sam Altman.

However, the plan faces significant skepticism. Critics point to the potential for regulatory capture, noting that a body funded and staffed by the industry might struggle to maintain total independence. Furthermore, there is concern that the proposal would impose disproportionate compliance costs on smaller startups and open-source developers, effectively cementing the dominance of well-resourced incumbents. Others highlighted a jurisdictional limitation: without participation from foreign powers such as China, a US-led watchdog might simply disadvantage domestic firms while failing to mitigate global risks.

What to Watch Next

As the debate over safety standards continues, the following points remain critical:

• Legislative Backing: Whether the U.S. Government will codify this voluntary model into law or maintain a purely advisory role.
• International Coordination: Efforts to align the proposed US-led coalition with existing frameworks, such as the European Union’s AI Act.
• Timeline: Hassabis reportedly hopes to have the organization operational before the end of the year.
• Enforcement Details: Clearer definitions regarding the legal consequences for labs that fail to pass safety benchmarks or refuse to pause model deployment.

For now, the proposal remains a vision rather than a finalized policy. It signals a shift in the business environment where top-tier labs are actively soliciting the very regulations they once avoided, though the path toward global consensus remains uncertain.