The “father of AlphaGo,” Demis Hassabis, CEO of Google DeepMind, has publicly proposed the establishment of a privately operated, U.S.-led AI standards body ahead of the era of artificial general intelligence (AGI). Hassabis emphasized the need for an independent organization dedicated to pre-deployment evaluation and national security risk verification of cutting-edge AI models, using the Financial Industry Regulatory Authority (FINRA) as a model.
On the 14th (local time), Hassabis posted a lengthy article on his X (formerly Twitter) account titled “A Framework for Frontier AI and the Dawn of a New Era.” He stated, “We are only a few years away from the emergence of AGI,” and predicted that “the scale of this technology’s impact is unprecedented, likely unfolding 10 times larger and faster than the Industrial Revolution.” He argued that, given its economic and technological standing, the United States is in the most advantageous position to establish a regulatory framework and should maintain a stance of “cautious optimism.”
The specific blueprint proposed by Hassabis mirrors FINRA, a private self-regulatory organization that oversees Wall Street under the supervision of the U.S. Securities and Exchange Commission (SEC). The new body would be tasked with evaluating AI models and collaborating with federal agencies to test models in areas related to national security, such as cybersecurity and biological risks. The organization would be funded by the AI industry but operate independently with government support.
A phased regulatory approach was also detailed. Initially, cutting-edge AI labs would voluntarily submit their models to the standards body for review 30 days before release. Once the effectiveness and robustness of this evaluation process are proven, mandatory pre-approval for distribution within the U.S. market would be required. “The strength of this approach is that it focuses on the technology while supporting innovation and encouraging responsible behavior,” Hassabis explained. “It can adapt as the greatest risks are identified, keeping pace with the speed of technological advancement, and the regulatory intensity can be increased depending on the severity of the situation.”
This proposal is interpreted as an attempt to address the limitations revealed in recent ad-hoc reviews of AI models conducted individually by the U.S. government. Previously, the U.S. Department of Commerce issued an “export ban” directive temporarily prohibiting the provision of services to foreign nationals for models like Anthropic’s Claude Mythos on security grounds. Additionally, OpenAI underwent a limited release process for its latest model, GPT-5.6, initially making it available only to certain government-approved institutions. These makeshift reviews have faced significant criticism due to a lack of technical expertise and opaque decision-making processes. Hassabis’s vision is to transfer these decisions to an independent body composed of industry experts.
The controversy surrounding AI regulation remains heated within both the tech industry and the government. White House AI advisor Sriram Krishnan has dismissed the idea, stating that an agency akin to an “FDA for AI” would not be created within the administration. Hassabis’s FINRA model is effectively presented as a realistic alternative to circumvent such opposition. He added that the body should be composed of representatives from the open-source community and industry technical experts, and could potentially outsource some evaluations to AI safety groups specializing in specific risk areas. He predicted that once U.S.-led standards are established, they would pave the way for international standards on AI.
Google CEO Sundar Pichai also immediately expressed his support. Quoting Hassabis’s post on his own X account, Pichai commented, “Well said” and “worth reading,” demonstrating that Google’s top management is speaking with one voice on AI safety regulation.
Meanwhile, Hassabis, who is from the United Kingdom, is the figure who created a turning point in global AI technology by developing the Go-playing AI “AlphaGo” in 2016. This year marks the 10th anniversary of the “match of the century” between AlphaGo and Lee Sedol, a 9-dan professional Go player, which Hassabis visited South Korea to commemorate in late April. His public emphasis on the imminent arrival of AGI and his proposal of a concrete regulatory framework appear to stem from a strong sense of urgency that institutional preparedness is woefully inadequate compared to the pace of technological advancement.