Bessent says Trump and Xi agreed humans must stay in military AI decisions
The Treasury Secretary called it a headline result of the Shenzhen summit, and said it has nothing to do with the race for AI.
On November 20, 2026, Treasury Secretary Scott Bessent said President Trump and China's President Xi Jinping had agreed that humans must stay involved in military decisions made with advanced AI. He said it on CNBC's Squawk on the Street, back from the trip to Asia where the two leaders met in Shenzhen.
Asked for the trip's biggest accomplishments, Bessent put it near the top, alongside trade deals and nuclear talks. CNBC posted an unofficial transcript of the interview, which is where every quote here comes from.
President Trump, and President Xi agreed that humans should always be involved in any military decisions that SI make, makes, and the SI should not be allowed to be autonomous for major military actions.
What Bessent described
The transcript calls the technology SI throughout. In September we covered how Washington had started calling AI super intelligence in its talks with Beijing, and Xi used his White House welcome that month to call for keeping AI under human control. Shenzhen was the leaders' third meeting this year, Bessent said, and a fourth is planned at the G20 in Doral.

Sara Eisen pushed back. The administration's line has been to beat China and keep advanced chips from it, she said, so how do you cooperate at the same time? Bessent split the two apart. The U.S. is ahead in the race and must stay ahead, he said, and attacks using SI come from Chinese models that have spread around the world.
both sides agree that no major military operation should be launched without a human being involved.
He didn't say what counts as a major military action, or whether anything was written down. The saved sources hold no Chinese statement on it, and no text of the agreement from either government.
How it compares with 2024
Washington and Beijing have been here once before, on a narrower point. When Joe Biden met Xi in Lima, Peru, the White House readout said the two leaders agreed on one line.
The two leaders affirmed the need to maintain human control over the decision to use nuclear weapons.
Bessent's version is broader on paper. It covers major military actions in general, not just a nuclear launch. If it holds, it's a promise about how two militaries use AI, which reaches well past the labs. As described, though, it's one official's account of a conversation, with no definition and no way to check it.
The rest of what he said about AI
On chips, Bessent said nothing changes. The H200 stays the threshold for what can be sold to China, and he said Chinese orders for it have been thin as the country builds its own chips. Jensen Huang didn't come up in the talks, he said. Treasury and the National Security Council's rule on overseas data centers renting compute to China means remote access has been curtailed, in his words.
He was asked about Senator Elizabeth Warren's call for safeguards on advanced AI after the Anthropic IPO. Bessent said the administration has had policies on the labs since Mythos came out in March, with pre-testing and notification when an exploit gets out of control. The labs, he said, are the country's shield.
He also made some news on money. Bessent said one of Anthropic's founders will make a large gift to Trump Accounts, the government's savings accounts for children, over the next few weeks. He wouldn't say which one.
The next scheduled meeting between Trump and Xi is the G20 at Doral.
More on AI policy
- Trump says American companies that build AI data centers abroad will face a 'tremendous tax'November 21, 2026
- House chairmen urge banning Chinese AI models from federal work, and Cursor's maker may drop themNovember 21, 2026
- European Commission confirms a cyber campaign run with a public AI tool, detected October 22November 19, 2026
- Whitesides, Moran and Beyer launch what they call Congress's first AI safety caucusNovember 19, 2026


