The most consequential AI agreement of the Trump-Xi summit was not signed on the South Lawn. It was hammered out three days earlier in a New York conference room, and neither leader spent much time talking about it on Thursday.
President Donald Trump opened Chinese President Xi Jinping's state visit to Washington on September 24 by making clear that the two superpowers would not be negotiating limits on artificial intelligence. Hours before the arrival ceremony, he posted on social media that superintelligence would be discussed but that he intended to leave the technology alone. “A big day with President Xi of China. Super Intelligence (SI) will be a big topic of discussion, but I want to leave it exactly where it is. That is China's position also,” Trump wrote, according to Bloomberg. He followed with a one-line summary of his regulatory philosophy: “Our guardrail is the DOJ!”
Xi, speaking in the Grand Foyer of the White House, struck a noticeably different tone. “China and the U.S. are leading nations in AI and we have both the capability and responsibility to develop and manage AI for good and ensure the development of AI is always under human control,” he said. The Chinese leader, on his first state visit to the United States in 11 years, also warned that the two sides “will both lose in confrontation.”
A Hotline, Not a Treaty
Beneath the pageantry, officials did produce something concrete. Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng held an eight-hour meeting in New York on September 21 and agreed to a US-proposed channel for notifying each other about AI incidents that could rise to the level of a national security threat. According to reporting summarized by AI Weekly and NBC News, the categories under discussion include runaway AI agents, AI-enabled cyberattacks and bioweapons development. Commentators quickly compared the mechanism to the Washington-Moscow hotline created after the Cuban Missile Crisis.
China's commerce ministry said Thursday that the two countries had held their first dialogue on AI, though the state news agency Xinhua gave the subject only brief mention in its readout. The two sides agreed to reconvene in Shenzhen in roughly two months to continue talks on AI safeguards.
The AI track was not the only deliverable. Within minutes of Xi's arrival at Joint Base Andrews, Bessent announced that Washington and Beijing had extended their trade truce by about two months, to January 10. Trump also touted expanded access to the Chinese market for American farmers under a newly created Board of Trade, and Xi announced that two pandas would arrive at Zoo Atlanta within days.
Who Was in the Room
The guest lists told their own story about how each side views the technology. Thursday evening's state dinner, only the second of Trump's current term, included American technology leaders such as OpenAI chief executive Sam Altman, with Reuters reporting that Nvidia's Jensen Huang, Apple's Tim Cook and Meta's Mark Zuckerberg were also expected. China's delegation, by contrast, conspicuously left out its frontier AI founders. DeepSeek's Liang Wenfeng and Moonshot AI's Yang Zhilin were absent, according to the South China Morning Post and the Seoul Economic Daily, while electric vehicle and battery firms such as BYD and CATL were more visible.
Analysts read the omission as a deliberate effort by Beijing to shield its most sensitive AI talent from exposure in the United States and to keep frontier model development off the negotiating table. It stood in contrast to Trump's May trip to Beijing, when executives including Huang and Cook flew aboard Air Force One.
Not everyone in Washington welcomed the red-carpet treatment. Senator Roger Wicker, the Republican chairman of the Senate Armed Services Committee, said he would have advised against the invitation. “His guest is a brutal, unelected and oppressive dictator who seeks to dominate his neighbors,” Wicker said this week. Tensions also surfaced when Bloomberg reported that China had taken possession of F-35 stealth aircraft parts diverted to Hong Kong.
Why It Matters
The summit crystallized an unusual inversion in global AI politics. The Chinese government, long criticized in the West for its approach to technology control, positioned itself as the champion of human oversight and shared responsibility. The American president, meanwhile, rejected guardrails outright, days after using a United Nations speech to dismiss calls for an international AI agreement and arguing that existing institutions can handle the risks. Trump has previously called warnings that AI could pose an existential threat a “hoax.”
Yet the incident notification channel suggests both governments see value in a narrow safety net even while refusing broader constraints. A crisis line for runaway agents or AI-driven cyberattacks does nothing to slow capability development, which is precisely why it was achievable. It lets Washington pursue unfettered competition while giving both sides a way to avoid misreading an AI accident as a deliberate attack. For AI labs on both sides of the Pacific, the key unanswered question is whether private companies will be expected to report incidents into the system, or whether it will cover only government-run systems.
The exclusion of DeepSeek and Moonshot also underscores how both governments now treat frontier AI talent as a strategic asset, not simply a commercial one.
What to Watch
The Shenzhen round, expected in about two months, is the real test. A written incident-notification protocol emerging from it would mark the first formal bilateral AI safety arrangement between the United States and China; a postponement or a vague communique would suggest Thursday's warm words were mostly ceremony. Watch too for whether Washington ties the channel to export controls on advanced chips, which Beijing continues to protest, and whether the January 10 trade truce deadline becomes leverage in the AI talks.
“Our guardrail is the DOJ!”— Donald Trump, President, United States