U.S. Proposes Notification Channel for AI Incidents — China Has Confirmed the Dialogue, Not the Mechanism

U.S. Proposes Notification Channel for AI Incidents — China Has Confirmed the Dialogue, Not the Mechanism

When Treasury Secretary Scott Bessent announced a new U.S.-China AI dialogue on September 21, 2026, after talks in New York, the only concrete proposal on the table was a notification mechanism — not binding limits. The meeting, held September 20–21 ahead of the summit between Donald Trump and Xi Jinping on September 24, ended with agreement on the dialogue itself, while the notification mechanism for AI incidents with national security implications remains only a U.S. proposal. Chinese authorities have confirmed that the dialogue took place, but have not commented on the details.

What Was Actually Agreed

According to Bloomberg News, quoting Bessent directly, the Treasury Secretary said the parties had agreed to establish what he called a "U.S.-China AI dialogue" — a dialogue intended to develop a shared understanding of the technology's benefits and threats. In addition, Washington proposed a notification mechanism covering AI incidents with national security implications (Charlotte Observer/Bloomberg).

The distinction is worth emphasizing: the agreement concerns the dialogue. The notification mechanism was proposed by the United States, and no source confirms that China has signed on to it. The New York Times' DealBook, which covered the meeting, described the proposal as a kind of "red telephone" for the AI age, in which the countries could alert each other to national security issues related to the technology (NYT DealBook). Bessent framed the proposal as a matter of transparency: according to the NYT, he said transparency "between the world's largest and second-largest AI power is very important."

Bessent himself called the meeting "very successful" in comments to the press after the talks, which, according to the New York Post, focused on trade and AI (NY Post).

What the Mechanism Actually Covers — and What It Doesn't

Bloomberg News' analysis is clear: the initiative for now appears to be about risk management, not binding constraints on AI development. That comes despite calls from Anthropic co-founder Dario Amodei and other American tech leaders to slow the development of frontier models.

Concretely, this means the proposed mechanism would function as a communication channel: if one party perceives an AI-related incident with national security implications, there would be a way to notify the other party, rather than the matter surfacing as a surprise point of suspicion. NYT DealBook's "red telephone" image points to the Cold War model, when a direct line between the superpowers was meant to reduce the danger of misunderstandings.

What the mechanism does not do is set limits on what either party can build or deploy. No source in the available material suggests the dialogue will lead to arms-control-style agreements, capacity limits, or standardized rules.

What Bessent Says Comes Next

According to Financial Times, which Bloomberg News cites, the U.S. also wants to raise threats from non-state actors and concerns related to the use of AI in developing biological weapons. Bessent added that the parties will likely meet again in roughly two months. That is, for now, the most concrete timeline for follow-up — and even that is phrased as probable, not confirmed.

The Chinese Version Is Effectively Missing

The Chinese response, compared with the American one, is nearly silent. The state news agency Xinhua confirmed that officials held a dialogue on AI, but gave no further details (Charlotte Observer/Bloomberg). That means the Chinese government's own characterization of what the dialogue is meant to cover — and whether Beijing has actually accepted the notification mechanism — is effectively absent from the public record.

Three days after the New York meeting, in his opening remarks at the White House on September 24, Xi offered the fullest Chinese framing available. He said both countries have both the ability and the responsibility to develop and govern AI for the good, and that AI development must be kept "always under human control" (New Indian Express). That is a statement of principle, not a response to the American mechanism.

The Timeline: The New York Talks and the Summit After

It is tempting to read the New York negotiations as a breakthrough that also settled trade, but the chronology is more complicated. According to Financial Times, as reported by the Charlotte Observer, the New York talks did not produce agreement on extending the countries' trade truce. The two-month extension came only in the days around the summit itself on September 24. The AI dialogue and the trade extension thus belong to two different moments and two different decisions.

At the summit itself, AI was on the agenda. Trump had said beforehand that "superintelligence" would rank high on the agenda, but at the same time positioned himself as favoring the status quo: "I will leave it exactly where it is. That is also China's position," he wrote in a social media post cited by the Associated Press (Chicago Tribune/AP). That is a striking statement from the leader of one of the world's two AI powers: neither Trump nor Xi is signaling any desire to constrain technology development through this process.

Why Seasoned Observers Are Skeptical

Kurt Campbell, former Deputy Secretary of State and co-founder of The Asia Group, offered a dry assessment on Bloomberg TV of how far Beijing will go: "They're going to be reluctant to enter into anything that would constrain Chinese actions" (Charlotte Observer/Bloomberg). Campbell added that the AI discussions have not yet reached a stage where the parties are ready to act.

The background for the skepticism is concrete. Since Trump and Xi last met in May, AI has become a central point of friction between the countries. American officials, security agencies, and technology companies have accused Chinese companies of systematically distilling — that is, training their own models on the outputs of — American models. Beijing has rejected the accusations as groundless and has vowed to respond if Washington attacks Chinese AI companies. It is against this backdrop that a notification channel is meant to function — between two parties actively accusing each other of stealing technology.

What Remains Unresolved

Several things remain open. No official documents or complete statements from Bessent have been published; all reporting so far is secondary, and the Bloomberg and NYT DealBook coverage draws partly on the same briefings, so they should not be read as independent confirmation of each other. It is unclear exactly what the proposed notification mechanism is meant to cover in practice, who would operate it, and at what level. And it remains unanswered whether the September 24 summit itself produced concrete AI announcements beyond Xi's opening remarks and the trade decisions.

Bessent points toward a follow-up meeting in roughly two months. Campbell has pointed to possible later venues for further progress — including the APEC summit in Shenzhen in November and the G20 meeting in Florida in December — but these are presented in the sources as possibilities, not confirmed steps.

What is established so far, then, is modest: a formal dialogue between the world's two leading AI powers, an American proposal for a notification system for national security incidents, and a shared willingness to keep talking. If the goal was to slow or constrain AI development — as Amodei and others have urged — this meeting was not a step in that direction. If the goal was to keep the next AI-related incident with national security implications from escalating in an information vacuum, the parties now have, at minimum, a framework for notifying each other.