US Proposes AI Incident Alert System With China Ahead of Trump-Xi Summit

Treasury Secretary Scott Bessent proposed a bilateral alert system for national security-level AI incidents during talks with China's He Lifeng in New York.

The United States has proposed a formal notification system that would require Washington and Beijing to alert each other to artificial intelligence incidents that could threaten national security, Treasury Secretary Scott Bessent said over the weekend, according to Reuters and the Associated Press.

The proposal was tabled during economic meetings with Chinese Vice Premier He Lifeng in New York, held as the two governments raced to lock in deliverables before a planned meeting between President Donald Trump and President Xi Jinping later this week. US Trade Representative Jamieson Greer also took part in the discussions, according to multiple reports.

What the Proposal Would Do

According to details reported by Reuters and confirmed by subsequent coverage in the Washington Post and the New York Times, the mechanism would create a dedicated communication channel between the two governments for AI-related incidents that rise to a national security threshold.

The idea mirrors existing risk-reduction arrangements in other domains, where adversaries commit to notifying each other about events that could otherwise be misread as hostile acts. In the AI context, an incident could range from a frontier model failure with cross-border consequences to a major AI-enabled cyber operation whose origins are unclear.

Bessent framed the proposal as a move from opacity toward transparency, arguing that when two countries dominate global AI development, silence between them creates more risk than communication, according to reporting on the talks.

Talks Before the Summit

The New York meetings were the latest round in a burst of US-China economic diplomacy, and they carried obvious weight beyond the negotiating table: President Trump has planned a White House state dinner for later this week that, according to prior reporting, is expected to draw top AI executives including Sam Altman, Jensen Huang, Elon Musk and Sundar Pichai. The Financial Times reported that the two sides agreed to a dialogue on AI ahead of the Trump-Xi meeting, and Chinese state-linked outlets described the arrangement as an AI incident hotline paired with revived trade talks.

The negotiations covered more than AI. Reports from the talks indicate the two delegations discussed potential tariff reductions on certain goods and the operationalization of a "Board of Trade" framework that had been outlined in earlier rounds. But the AI notification proposal stands out as the most concrete security-adjacent measure to emerge from the weekend, and it gives both leaders something substantive to discuss when they meet.

For Bessent, the proposal is also a continuation of a noticeable shift in tone. Earlier in the week, he said the United States was open to discussing shared AI risks with China, language that contrasted sharply with President Trump's repeated characterization of climate-style catastrophe claims about AI as a "hoax."

The Hard Part: Defining an Incident

Turning the proposal into a working system faces familiar obstacles. The first is definitional: neither government has said what counts as a "national security-level AI incident," and the two countries have sharply different views on what constitutes a technology security threat.

The second is trust. The United States has spent recent years tightening export controls on advanced chips and AI technology, while China has invested heavily in building parallel AI infrastructure designed to reduce its dependence on American technology. A notification channel does not dissolve that competitive dynamic; it simply manages the risks that come with it. There is also a sequencing problem: the proposal emerged from an economic dialogue led by treasury and trade officials, while AI risk policy in the US spans multiple agencies with their own jurisdictions — a structure that makes it unclear who would actually receive, verify and respond to an incoming alert on either side.

Even so, officials appear to see the mechanism as low-cost insurance. An alert system creates no obligation to share model weights, restrict exports or coordinate research. It asks only that each side pick up the phone when something goes wrong — an unusually modest ask by the standards of US-China technology diplomacy, and possibly the reason it advanced as far as it did.

Policy researchers have been pushing for exactly this kind of arrangement. Think tanks, including the Institute for AI Policy and Strategy, have argued that ad hoc communication about critical AI incidents leaves too much room for misinterpretation, and that notification mechanisms should focus on clarifying responsibilities and threats without requiring either side to reveal proprietary capabilities.

Why It Matters

If implemented, the system would be one of the first formal, government-to-government AI risk channels between the world's two leading AI powers. Existing AI safety commitments, including the frontier AI safety pledges signed at AI summits in recent years, are largely voluntary and industry-focused. A bilateral state-level mechanism would be a different category of commitment entirely.

Market reaction so far has been muted, with no immediate implications reported for AI-related trading. That could change depending on what comes out of the Trump-Xi meeting. If the two leaders endorse the notification concept and attach timelines or enforcement details, it could begin reshaping expectations around US-China AI cooperation. If the summit ends without agreement, the proposal joins the long list of US-China AI governance ideas that stalled at the announcement stage.

Either way, the weekend talks mark a shift: the conversation between Washington and Beijing on AI is no longer about whether to talk about risk, but about how the warning system would actually work.