According to the Wall Street Journal, sources said that a few weeks before stepping down as CEO of Google DeepMind, Demis Hassabis had discussions with U.S. government officials and other AI lab leaders about establishing an independent AI industry safety organization. As a co-founder of DeepMind and a pioneer in machine learning, Hassabis has increasingly devoted his time to promoting the creation of organizations that develop safety mechanisms and best practices for general artificial intelligence. When discussing this proposed institution with peers, he compared it to the International Atomic Energy Agency—a non-governmental oversight organization that promotes cooperation in nuclear energy.

Sources said that in early summer this year, Hassabis discussed all aspects of the plan with senior officials from the Trump administration, including U.S. Treasury Secretary Scott Bessette and one of President Trump's senior technology advisors, Michael Kratsios. A Google spokesperson said the company believes ensuring the safety of advanced AI models is "crucial for society," and Hassabis has a unique advantage in advancing this goal, stating that he will take on a new role at Google and strategically focus more on scientific breakthroughs and shaping the direction of AGI.

Introducing a "Frontier-Level" Labeling System, Modeled After the Financial Industry Regulatory Authority

In mid-July this year, Hassabis posted on X, suggesting that the United States should establish an industry-funded "standardizing body" to help assess the safety of AI models. He said that the institution could operate similarly to the Financial Industry Regulatory Authority, a self-regulatory organization established during the Great Depression to oversee brokerage firms, funded by the financial industry. According to his plan, federal agencies and national laboratories under the U.S. Department of Energy would collaborate with AI companies to test models in "fields related to national security" and establish a certification system, granting models that meet specific standards a "Frontier-Level" label.

Last week, Google announced major management changes at DeepMind, with Hassabis stepping down as CEO and being succeeded by his deputy, Corinna Kretzschmar, while Hassabis himself took on the roles of Chairman of DeepMind and Chief Scientist at Alphabet. This "succession plan" before his resignation clearly indicates preparations for the next chapter in AI governance—transitioning from leading the world's most advanced AI laboratory to attempting to build a safety governance framework for the entire industry. Hassabis is shifting from a technical leader to a systems designer, and if this proposed institution comes to fruition, it could have a profound impact on the global AI regulatory landscape.