WASHINGTON — Treasury Secretary Scott Bessent said on Friday that President Donald Trump and President Xi Jinping of China agreed in Shenzhen this week that humans must always be involved in military decisions made with artificial intelligence, and that the technology must not be allowed to act on its own in major military actions.
Bessent, who traveled with Trump, described the understanding on CNBC as one of the major achievements of the president's trip to Asia. "I think it was very important that President Trump and President Xi agreed that humans should always be involved in any military decisions that [AI] makes, and the [AI] should not be allowed to be autonomous for major military actions," he said. Bessent used the term "SI," short for super intelligence, the Trump administration's preferred name for the technology.
He said the agreement stood apart from the race between the two countries to build the most powerful AI systems. "But there's nothing to do with competition when it says that both sides agree that no major military operation should be launched without a human being involved," he said.
Trump used his own words at a White House news conference on Friday. "President Xi and I agree that we're going to do some joint guardrails, the two of us, because there really is no third place," he said, arguing that the United States is leading in the technology, China is second and no other country is close. On Thursday, in an interview with Fox News, Trump had said the two leaders both wanted AI safety rules and would work together on reasonable ones.
No written text of the understanding has been published by either government. It was not part of the four-point declaration that the United States, China and Russia issued in Shenzhen on Thursday, and neither side has announced definitions, verification steps or a timetable. What the public knows about the military commitment comes from the American side, from Bessent's interview and Trump's remarks.
China's public words back the general principle but do not describe a military agreement. The official Chinese readout of Thursday's meeting, published by the Xinhua news agency, said Xi told Trump that the two countries should "strengthen communication and coordination to prevent the safety risks of AI and ensure that AI is always under human control." China's embassy in Washington used nearly the same words in a statement on Thursday. Beijing has not publicly confirmed the military details Bessent described.
The idea has a history. In November 2024, President Joe Biden and Xi met in Lima, Peru, and "affirmed the need to maintain human control over the decision to use nuclear weapons," according to the White House account of that meeting. The two governments had opened a formal dialogue on AI risks in Geneva earlier that year. What Bessent described goes wider than the 2024 statement, which dealt with nuclear weapons. His version covers military decisions in general.
American policy already requires a human role at home. A Pentagon directive updated in 2023 says autonomous and semi-autonomous weapon systems must be designed so commanders and operators can exercise "appropriate levels of human judgment over the use of force." In 2023 the United States also launched a political declaration on responsible military use of AI that dozens of countries have endorsed. China has not joined it.
In the same CNBC interview, Bessent said Nvidia's H200 chip remains the most advanced chip China can buy under American licenses, and that Chinese buyers have shown little appetite for it. "I think what we have seen is a paucity of orders from the Chinese on the H200s as they try to build out their own chip stack," he said. He said the spying allegations involving former Nvidia chief executive Jensen Huang did not come up in the meeting between Trump and Xi. The Commerce Department moved on Monday to further restrict China's access to advanced chips, including through data centers in other countries where Chinese companies rent computing power.
Nothing announced this week creates inspections, penalties or any way to check compliance, and analysts who spoke before the summit expected no binding safety standards to emerge from it in the short term. Whether the understanding becomes more than words will depend on whether the two governments put it in writing and define what it covers.
Reporting note: This story is based on Reuters reporting of Bessent's CNBC interview and of Trump's news conference and Fox News interview, on the Chinese government's published readout of the Shenzhen meeting, and on earlier White House, Pentagon and State Department records on military AI policy.
