United States Treasury Secretary Scott Bessent said artificial intelligence companies should be held legally responsible for their models' actions.

"These labs need to take responsibility for themselves," Bessent said Monday during an interview on CNBC’s Squawk Box. "They can slow down any time they want to."

Bessent cited recent industry warnings that advanced AI could potentially wipe out humanity, while companies are seeking protection from the financial consequences of catastrophic events.

"If you think about it in terms of the valuation, what did they try to do last week?" Bessent said. "It was, 'Well, there’s a 10% chance we could destroy the world, but we want the government to give us a liability shield.'"

"That’s good business for them, bad business for the American people," he added.

Bessent agreed with MIT professor Daniel Huttenlocher that the humans in charge of these models, not the AI agents themselves, should be held accountable if their technology causes harm.

He pointed to the recent incident where OpenAI models managed to escape their test environment and compromised parts of Hugging Face's systems, saying that the responsibility ultimately falls on OpenAI.

Bessent's comments come as President Donald Trump pushes back against calls to slow down AI development, as doing so would cede the U.S.'s lead to China. Last week, Trump called fears that AI could destroy humanity a "hoax," while saying that existing government powers are enough to hold companies responsible.

US-China talks

Bessent also revealed a new U.S.-China dialogue following talks with Chinese Vice Premier He Lifeng. The two countries are planning to establish an incident communications line and meet again near the end of the year.

Those discussions are expected to focus on outlining shared threats, including the aforementioned out-of-control agents, cyberattacks and the use of AI in developing chemical weapons.

Bessent said that China acknowledged the US' lead in AI, though he described Chinese open-source models as "very powerful."

"We’ve seen this push for frontier with the labs," Bessent said. "Now we need to see the frontier on defense and resilience."