Treasury Secretary Bessent Proposes US-China AI Notification Mechanism

The United States has proposed building a mechanism through which it and China would alert each other to AI incidents. US Treasury Secretary Scott Bessent told reporters after a full day of talks with the Chinese delegation in New York on Sunday, September 20, that the two sides had discussed a new US-China AI dialogue containing an arrangement for notifying each other of AI-related incidents that could affect national security. It is a proposal headed for the summit between President Donald Trump and President Xi Jinping in Washington this week.

The talks were held at JPMorgan Chase headquarters in New York. The Chinese delegation was led by Vice Premier He Lifeng and included China's chief trade negotiator Li Chenggang, while the US side was represented by Bessent and US Trade Representative Jamieson Greer. Bessent said, "We think that just like with any cross-border activity, that moving from opaque to more transparency between the number one and the number two AI powers in the world is very important", adding that the two sides had agreed to meet again on the AI issue.

The Chinese record reads at a different temperature. The official Chinese account of the talks that METAL checked says the two sides held candid, in-depth and constructive exchanges on economic and trade issues of common concern and on implementing the consensus reached in earlier talks, guided by the principles of mutual respect, peaceful coexistence and win-win cooperation. On AI, a single sentence saying the two sides held dialogues on AI-related issues is all there is. The notification mechanism the US put forward does not appear in that account.

The scope of the proposal is drawn narrowly. According to reports, what would be notified are AI-related incidents that rise to a national security level, described as an arrangement covering common goals and common threats. Greer drew a line, saying US export controls on advanced AI chips and semiconductor manufacturing equipment were not on the agenda for these AI mechanism talks. The very measures that have held back Chinese AI companies are being kept off the table.

This is not the first time the two leaders have raised AI consultations. Trump and Xi discussed the possibility of consultations on AI development in Beijing in May, but that forum was never formalised. The current proposal is an attempt to revive, in institutional form, a conversation that stalled there. METAL has reported that China's minister of state security publicly named six AI risks, so the vocabulary for counting risks already exists on both sides.

Plenty besides AI was on the table. The two sides agreed to actually operationalise the Board of Trade announced in Beijing in May, and to explore possible tariff reductions on non-sensitive goods. According to Greer, candidates on the Chinese side include consumer goods and low-tech items, while on the US side energy, agricultural goods and medical devices came up. "There is a world where you have a relatively small subset of American and Chinese goods that can be traded in a balanced way that are non-sensitive", he said.

There is a clock running. The trade truce struck last year expires on November 10, and among the commitments that emerged from the May talks in Beijing, an additional $17 billion a year in US agricultural purchases and orders for more than 200 Boeing aircraft have yet to show confirmed progress. US officials say the flow of critical minerals to American companies also falls short of what Beijing promised to ease. The AI notification mechanism is a new item stacked on top of that list.

Expectations for how far AI will get at the summit are largely sceptical. Observers point to deep distrust between the two countries as a reason a major agreement is unlikely, and China has read calls to slow the pace of AI development as aimed at its own progress. Anna Ashton, a China trade analyst and founder of Ashton Intelligence, said before the talks that "status quo is probably both sides' general best expectation". Greer said, "These are the two most powerful countries on earth, whether it's in trade or AI, and it's imperative that they're able to work together."

The air around the summit does not blow in only one direction either. Trump has dismissed industry warnings that AI could become an existential threat within a few years as a hoax, while agreeing with Anthropic CEO Dario Amodei that the US must not let China pull ahead in the AI race. He Lifeng brought a delegation of Chinese companies with him on this trip, mirroring the group of US chief executives Trump took to Beijing in May. In the same week, US auto industry groups asked that the effective ban on Chinese vehicle sales be maintained on national security grounds.

There are voices pushing the demands higher. Representative Ro Khanna, the ranking Democrat on the House Select Committee on Strategic Competition between the United States and the Chinese Communist Party, said on a Sunday program that the two leaders must produce an international agreement covering AI safety: "The first ask is of Trump and Xi Jinping to come up with an international agreement, and two of the core principles should be one banning any self-improving AI, recursive self-improving AI." The second, he added, is safeguards so that AI cannot be misused for biological or nuclear weapons.

The notification mechanism is a proposal, not an agreement. It goes to the summit table present in one side's account and absent from the other's. Even so, this is the first time the idea of settling in advance on a number to call when something goes wrong has reached the two countries' formal agenda. What Thursday will reveal is not the size of an agreement but whether that number gets written down.