The head of Artificial Intelligence (AI) at Google called on Tuesday for the United States to create a new organization that would test the most powerful AI systems before they are made available to the general public.
In a piece published on Substack, Demis Hassabis, who leads DeepMind, argued that an AI capable of matching the full range of human intellectual capabilities will be available within a few years.
The scientist, who won the Nobel Prize in Chemistry in 2024 for his protein research, heads Google’s division responsible for all of the group’s AI tools, competing with OpenAI and Anthropic in the race for the most powerful systems.
His proposal comes as executives at his competitors have already urged stronger oversight by the U.S. government over the technology, a few weeks after tensions between the government and Anthrop ic over the lack of clear rules in the sector.
The Commerce Department ordered in June that the company suspend access by all foreign customers or employees to its Claude Fable 5 and Mythos 5 models, citing national security risks.
The company had to make both models completely inaccessible for two weeks, before gradually restoring access.
According to Demis Hassabis, the regulator could be modeled on the Financial Industry Regulatory Authority (FINRA), a private regulator that oversees Wall Street brokerage houses under federal supervision.
The new entity would be formed from the best technical experts and initially funded by the AI industry, with the mission of developing the formal framework that a model must meet before it is brought to market.
The rules would apply to any model deemed sufficiently powerful to be categorized in the “frontier class,” whether built in the United States or abroad, Mr. Hassabis said, hoping that foreign companies would join the supervisor once it is established.
Besides the United States, China is among the countries where AI model development is the most advanced, Chinese companies offering products considered competitive but at lower cost.
“When there is such a degree of uncertainty and the stakes are high, moving forward with prudent optimism is the right strategy,” he explained.
Demis Hassabis is not the first to propose such an approach: Anthropic CEO Dario Amodei also called for more regulation by the government in early June and defended the idea of a federal agency that could test models before deployment.
