Plan modeled on nuclear watchdog and brokerage regulator
Alphabet chief scientist Demis Hassabis held discussions with government officials and leaders of other artificial-intelligence labs about forming a new independent industry safety entity in the weeks before he relinquished his role as chief executive of Google DeepMind, according to people familiar with the matter.
Hassabis, who co-founded Google DeepMind and is considered a pioneer in machine learning, has been spending increasing amounts of his time advancing the idea of an organization that would codify safety guardrails and best practices for developing artificial general intelligence, those people said. In discussions with peers, he has likened the proposed entity to the International Atomic Energy Agency, a nongovernmental watchdog that promotes cooperation on nuclear energy.
Earlier in the summer, Hassabis discussed aspects of his plan with high-ranking Trump administration officials, including Treasury Secretary Scott Bessent and Michael Kratsios, one of President Trump’s top tech advisers, people familiar with the matter said. Bessent has helped shape the administration’s AI strategy, warning that recent models like Anthropic’s Mythos pose cyber threats to the financial system.
In mid-July, Hassabis published an essay on X arguing that the U.S. should create an industry-funded “Standards Body” to help assess the safety of AI models. He said the entity would operate like the Financial Industry Regulatory Authority, a self-regulatory body established during the Great Depression that oversees brokerage firms and is funded by the sector. Under his plan, federal agencies and the Energy Department’s national laboratories would work with AI companies to test models “in areas relevant to national security” and would establish qualifications that would label a model “Frontier-class” if it met certain standards.
Last week, Google announced a management shake-up at DeepMind, its AI research unit, that saw Hassabis step down as CEO to be replaced by his deputy, Koray Kavukcuoglu. Hassabis was named to the joint roles of chairman of DeepMind and chief scientist of Alphabet. In a memo, Hassabis wrote: “I’ve been working towards AGI my whole life and now, like many of you, I feel it is close at hand.” He said he decided that now is the right time to hand over his day-to-day operational responsibilities at GDM “so that I have the time and space to focus on the big picture and help influence what is to come to the best of my ability.”
Jeff Dean, the company’s longtime chief scientist and a co-founder of DeepMind who was key to the creation of several of Google’s signature products, stepped down to found an AI-research startup. In a blog post last week, Alphabet CEO Sundar Pichai wrote that he and Hassabis “have been long discussing a role that allows him to put his full attention on actively shaping the future of AGI.”
Hassabis didn’t respond to requests for comment. A Google spokesperson said the company sees ensuring the safety of advanced AI models as “vitally important to society” and that Hassabis is uniquely positioned to work toward that goal. “Demis chose to step into a new role at Google and is focusing his efforts more strategically on scientific breakthroughs and shaping the future of AGI,” the spokesperson said.
Major AI labs are pursuing AGI, or computers capable of humanlike intelligence, and many of the top minds in tech, including leaders of the companies themselves, have expressed fears that the spread of the technology will disrupt society or that its power will be difficult to control. Recent instances in which AI systems carried out cyberattacks have added to widespread anxieties that the development of the technology is moving too fast and without enough regulation. The Trump administration has increased scrutiny of the security risks posed by AI but kept model evaluations for the most powerful tools voluntary.
Leaders at both OpenAI and Anthropic have called for establishing new mechanisms that would allow AI companies to coordinate on safety matters. OpenAI CEO Sam Altman last month proposed “a U.S.-led international forum” to determine standards and provide governance over labs, while Anthropic endorsed a global agreement to slow development and verify compliance.
Hassabis, a Nobel laureate, discussed the need for the U.S. to lead a globally coordinated effort to establish AI standards at a June Group of Seven summit in France that included Trump, other world leaders and AI executives including Altman and Anthropic CEO Dario Amodei.
Critics of the White House’s approach and of Hassabis’s idea say that voluntary efforts and self-regulatory concepts aren’t adequate to address the security and economic threats posed by AI. Many lawmakers are calling for mandatory model reviews and increased government oversight.