US China AI cooperation is entering a new and more candid phase, one in which Washington and Beijing will talk more often about the risks of advanced artificial intelligence while competing harder than ever to lead it. That was the message from Treasury Secretary Scott Bessent on Tuesday, when he told an audience in New York that the two countries should work together where their interests overlap — and be honest about where they do not.
Bessent was speaking in prepared remarks to the National Committee on US-China Relations, a forum that has long brought together business leaders, former officials and scholars who follow the relationship. His speech arrived at a sensitive moment: just days before he travels to China, and a week before President Donald Trump is due to meet President Xi Jinping again on the sidelines of the Asia-Pacific Economic Cooperation summit in Shenzhen.
The central idea of US China AI cooperation was balance. The United States and China, Bessent argued, do not need to agree on everything to agree on guardrails. They can exchange information, test powerful new systems more carefully and keep a direct line open when something goes wrong — even as they race to build the most capable models in the world.
US China AI Cooperation: What Bessent Actually Proposed
The most concrete proposal for US China AI cooperation in Bessent’s remarks was a new bilateral dialogue on superintelligence, often shortened to SI. The dialogue is intended to focus on the most advanced forms of AI, the systems that could one day outperform people across a wide range of tasks and create risks that neither country can manage alone.
According to the plan outlined by Bessent, the two sides will share best practices for evaluating and testing advanced models before they are released to the public. That means comparing notes on how to probe a model for dangerous capabilities, how to judge whether it is ready for wider use, and how to handle weaknesses found during testing.
They will also keep human judgment in command. In practice, that principle is meant to ensure that people — not automated systems — remain responsible for the most consequential decisions, particularly where AI could affect safety, security or the stability of critical systems.
A third element is a direct channel for alerting each other to serious incidents. If a powerful model behaves in an unexpected or dangerous way, or if a major AI-related incident occurs, the two governments would have an agreed route for raising the alarm quickly, rather than learning about it through rumours, markets or the media.
Bessent said the dialogue was meant to guard against shared risks and to prevent a misunderstanding between our two systems — a phrase that captures the fear that, without communication, a technical failure or a misread signal could escalate into a political crisis.
Clear-Eyed, Not Naive: Competition Will Continue
For all the talk of US China AI cooperation, Bessent was careful not to sound naive. He warned that the United States should not mistake a working dialogue for a convergence of values or interests.
“As we do so, we should harbor no illusions about the differences between the United States and China… our interests will not always coincide,” he said.
That sentence is likely to be the most quoted line from the speech, because it sums up the administration’s approach: cooperate on safety, compete on capability. Washington still wants to preserve its lead in advanced chips, frontier models and the infrastructure that powers them. Beijing still wants to close that gap and build a world-class AI industry of its own.
The alert system at the heart of the new dialogue was first agreed by Trump and Xi at their September meeting at the White House, according to reports of that encounter. Trump was said to be reluctant to accept new guardrails that might slow American companies down, a reminder that even safety measures are negotiated through the lens of national advantage.
The same tension runs through domestic AI policy in the United States. At home, the White House has brought technology chief executives together to discuss safety, investment and innovation, a debate covered in our earlier report on the Trump White House AI Summit, where tech CEOs discussed AI safety and innovation. The Bessent speech extends that conversation beyond America’s borders, to the one rival whose choices matter most.
The Road to Shenzhen: APEC and the Trump-Xi Meeting
Bessent travels to China next week ahead of Trump’s third meeting this year with Xi, on the sidelines of the November 18-19 APEC summit in Shenzhen. The choice of venue is symbolic as well as practical. Shenzhen is one of China’s great technology hubs, home to hardware makers, software firms and research labs that sit at the centre of the country’s AI ambitions.
The meeting is expected to advance the superintelligence dialogue and give both leaders a chance to test whether the September agreement on an AI alert system can be turned into working arrangements. Officials on both sides will be watching for signs that the dialogue has a timetable, named participants and a way of reporting back — the details that separate a declaration from a process.
A fourth Trump-Xi summit is expected in December, at the US-hosted G20 meeting in Miami. That gathering would put trade, technology and AI back on the agenda within weeks of the Shenzhen meeting, giving negotiators little time to let momentum fade.
Trade, Tariffs and Farming: The Wider Bargain
Bessent did not present AI as a standalone issue. He placed it inside a broader economic relationship that remains both deeply connected and deeply strained.
He reached back to 1784, to the first American trade voyage to China, to argue that commerce between the two countries is not new — and that it can still be managed to mutual benefit. Under the arrangements he described, a Board of Trade will determine mutual tariff cuts on $30 billion of non-sensitive goods in each direction, alongside China’s pledged purchases of American farm products.
A separate Board of Investment will cover market access for American firms operating in China, a long-standing complaint for US companies that say they face restrictions, uneven enforcement and pressure to share technology.
On the trade deficit, Bessent was blunt: the US deficit with China remains a source of friction. But he said the objective was to de-risk, not to decouple — to reduce dangerous dependencies without cutting off a trading relationship that still supports jobs, farms and factories on both sides.
That language matters for US China AI cooperation as well. De-risking in technology means protecting the most sensitive capabilities while allowing ordinary commerce, research and communication to continue. It is a narrower goal than full separation, and a harder one to define in practice.
What Bessent Did Not Say
Notably, Bessent made no mention of the investigation into Nvidia chief executive Jensen Huang reported by The New York Times, which said the company’s board was examining an affair with a businesswoman suspected by US security services of being a Chinese intelligence asset. China has called the accusations groundless, and Nvidia has said its board has no reason to believe secrets were compromised.
The omission was striking precisely because the story touches on the same anxieties that run through the AI debate: trust, technology transfer and the movement of people and information between the world’s two largest economies. By leaving it aside, Bessent kept the focus on government-to-government channels, rather than on a single company or individual.
Whether that separation can hold is an open question. In Washington, AI policy, export controls and corporate governance are increasingly discussed together. In Beijing, the same issues are viewed through the lens of sovereignty and development. A dialogue on superintelligence will have to operate in that charged atmosphere.
What to Watch Next
The test for US China AI cooperation will come in the next few weeks. First, in China, where his visit will show whether Beijing is ready to move from general agreement to specific commitments on testing, human oversight and incident alerts. Second, in Shenzhen, where Trump and Xi will decide how much political weight to give the new dialogue. And third, in December in Miami, where a fourth summit could either lock in progress or expose its limits.
For now, the message from New York is deliberately two-sided. The United States and China can, and should, talk about the dangers of the most powerful AI systems. They can share methods, keep people in charge and pick up the phone when something goes wrong. But they will still compete — for talent, for chips, for markets and for the right to set the standards that govern the technology.
That is what staying clear-eyed means for US China AI cooperation: cooperation without illusions, and competition without losing the ability to communicate. As Reuters reported, the coming weeks will show whether both capitals are prepared to live with that balance — or whether rivalry will once again crowd out the guardrails.