Demis Hassabis, CEO of Google DeepMind, called on Thursday for the creation of a US-led global AI standards organization to oversee the safe deployment of frontier AI models and manage associated national security risks [1, 2, 3]. He proposed the body be modeled after the Financial Industry Regulatory Authority (FINRA), functioning either as a federally overseen public-private partnership or a self-regulatory organization [1, 2, 3, 4].
Under the plan, AI labs would voluntarily share their frontier models with the standards body for up to 30 days before public release, allowing technical review and security assessments [1, 3, 4]. Hassabis said the entity would have authority to coordinate a coordinated slowdown of AI model rollouts if risks are deemed too high [1, 4].
Funding would likely come from industry contributions, while staffing would include independent technical experts and representatives from the open-source community [1, 2, 3, 4, 5]. The watchdog would also collaborate closely with federal agencies and national labs to perform security testing [5].
Hassabis stressed the urgency of such a body given rapidly advancing AI capabilities approaching artificial general intelligence (AGI), which he estimates may arrive by around 2030 [1, 6]. He said, "We’ve already seen the challenges frontier models pose for cybersecurity, and other threats including nuclear and bio risks may soon emerge as capabilities continue to advance" [2]. He added, "What we collectively do now will determine how the next phase of civilisation unfolds" [6].
The proposal has been extensively briefed to the Trump administration, European officials, and other AI labs, reportedly receiving positive responses [1, 2, 4]. Industry leaders including Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman also support a US-led AI standards body [2, 4]. The call came amid recent US government export controls and rollout limits on AI models due to national security concerns [2, 6].
Participating AI providers would be encouraged to adopt best practices such as comprehensive model documentation, strong internal cybersecurity, personnel review processes, and investment in security research [5].
Some analysts caution that a US government-led approach could raise geopolitical tensions and that self-regulation by the industry may face conflicts between shareholder interests and the public good [5]. However, Hassabis and supporters argue a technical, federally overseen model can balance innovation with risk mitigation [1, 2, 3, 4]. They emphasize a rapid though imperfect standard is preferable to slow inaction as AGI nears [5].
Hassabis hopes the new AI watchdog organization will be operational before the end of 2026 [1, 4].
