Demis Hassabis, CEO of Google DeepMind and Nobel laureate, wrote on the X social platform on Tuesday that the United States should initiate the creation of a standards organization to supervise new artificial intelligence models and assess national-security risks, including cyber and biological threats. He stressed that "urgent action" is needed to manage risks associated with Artificial General Intelligence (AGI).
AGI refers to a level of capability at which artificial intelligence reaches or exceeds human intellectual abilities. Hassabis warned that "we can already see the challenges that top-tier models pose to cybersecurity, and as their capabilities evolve further, other threats may soon appear, including nuclear and biological risks."
Proposed institutional design: a FINRA-style regulator
Hassabis proposed a public–private partnership operating under federal oversight, modeled after the Financial Industry Regulatory Authority (FINRA). Under his proposal, the new AI oversight body would have a board including independent senior technical experts and representatives of the open-source community. (FINRA is the U.S. self-regulatory organization that oversees broker-dealers and securities markets.)
He said the organization would require substantial funding to hire world-class staff and to secure the compute capacity needed for large-scale model testing; he expects funding would largely come from industry participants.
Voluntary start with the possibility of mandatory review
According to the proposal, leading AI companies would initially submit new models to the body voluntarily, up to 30 days before public release. If the review regime proved effective, such review could later become a mandatory pre-market condition for introduction on the U.S. market.
Review processes would pay particular attention to so-called agentic AI systems. The checks would include whether models attempt to circumvent safety constraints, exhibit deceptive behavior, and whether they implement best practices such as digital watermarking of AI-generated images or producing interpretable output tokens that help understand model reasoning.
Industry momentum and regulatory friction
Hassabis's suggestion came a month after, according to CNBC sources, he and Anthropic CEO Dario Amodei advocated for a U.S.-led coalition on AI rules and standards at a G7 summit that included U.S. President Donald Trump, tech company leaders and other heads of state. Earlier in the month, Sam Altman, CEO of OpenAI, also urged the establishment of a similar body in the Financial Times.
Despite growing support among industry leaders for an AI oversight organization, regulation of the most advanced AI models remains contentious between governments and private companies. In recent weeks, Anthropic engaged in extended talks with U.S. officials after the Trump administration imposed temporary export controls on one of its advanced models. OpenAI also faced government pushback when U.S. authorities initially asked the company to limit deployment of a new model.
Hassabis argues that the United States, given its economic and technological strength, is well positioned to lead the creation of a comprehensive AI regulatory framework.
Geopolitical angle: growing US–China competition
Calls for tighter regulation come amid intensifying competition between the United States and China in AI development and deployment. Chinese firms, including DeepSeek and Z.ai, have released models that some experts consider serious competitors to leading U.S. systems such as those from Anthropic and OpenAI. These alternatives are gaining traction among U.S. companies, partly because AI usage costs are rising.
As a result, U.S. lawmakers are examining ways to curb the spread of Chinese AI models within American firms; the U.S. State Department told CNBC that this trend is a "matter of serious concern."
