The US and China Want an AI Hotline. Here Is Why That Is Harder Than It Sounds.

The US and China Want an AI "Hotline." Here Is Why That Is Harder Than It Sounds.

Treasury Secretary Scott Bessent proposed a new notification system for artificial intelligence incidents that could affect national security, framing it as a way to bring transparency between the world's two largest AI powers.

The idea of a hotline between the United States and China is not new. During the Cold War, the red phone connected Washington and Moscow so that a nuclear launch would not unfold in silence. It was simple. One side launches missiles, the other side calls and asks what is happening. The problem was never the technology. It was the politics of picking up the phone.

Now Treasury Secretary Scott Bessent wants something similar for artificial intelligence. On Sunday, after talks with Chinese Vice Premier He Lifeng in New York, Bessent told reporters that the United States has proposed a "notification mechanism" for AI incidents that could threaten national security. The announcement comes just days before President Trump is scheduled to host Chinese President Xi Jinping at the White House for a state dinner on Thursday, September 24. The symbolism is hard to miss. The world's two largest economies, also its two most aggressive AI developers, are trying to build guardrails around a technology that moves faster than diplomacy.

Bessent described the proposal in deliberately broad terms. "We want a shared vision of common goals and common threats," he said. "We think that just like any cross-border activity, that moving from opaque to more transparency between the No. 1 and the No. 2 AI powers in the world is very important."

What he did not describe, because nobody really knows yet, is what an "AI incident" actually means in practice.

The Problem With Defining an "AI Incident"

A nuclear launch is unambiguous. A missile leaves a silo, satellites detect the heat signature, and within minutes the chain of command knows something has happened. There is no ambiguity about whether the event is worth reporting.

AI does not work that way. An autonomous trading algorithm could trigger a flash crash and wipe out billions before a human notices. A compromised language model could generate convincing phishing campaigns at scale. A military targeting system could misidentify civilian infrastructure as a threat. Are these "AI incidents" that rise to the level of national security? Who decides? And if the system that caused the problem is commercially owned, not government-run, does the obligation to report still apply?

These are not hypothetical edge cases. They are the everyday reality of a world where the most powerful AI systems are developed by private companies, deployed across global cloud infrastructure, and integrated into everything from supply chains to battlefield targeting. The boundary between civilian and military AI has never been blurrier, and neither Washington nor Beijing has shown much appetite for drawing a clear line.

The Business Dinner Behind the Diplomacy

If the proposal itself was vague, the guest list for Thursday's state dinner was not. According to reports from Bloomberg, the dinner will include Jensen Huang of Nvidia, Sam Altman of OpenAI, and Cristiano Amon of Qualcomm. These are not diplomats. They are the CEOs of companies that collectively control most of the world's AI training infrastructure, frontier model development, and mobile chip architecture.

Their presence signals something important. The Trump administration is not treating AI as a purely technical or military issue. It is treating it as an economic rivalry where American corporate dominance is a strategic asset. The same week Bessent proposed transparency with China, his administration is also hosting the very executives whose products give the United States its claimed edge in the field.

That tension runs through the entire proposal. You cannot simultaneously argue for transparency and for competitive advantage. If the United States shares too much about how it monitors and responds to AI incidents, it risks revealing capabilities China could adapt. If it shares too little, the notification mechanism becomes meaningless.

A Precedent, Not a Solution

George Chen, a partner at The Asia Group who focuses on digital policy, put the announcement in useful context. "The fact that both sides agreed to continue the dialogue is significant," he said. "The initial outcome, an AI risk notification mechanism, sets a precedent that other countries may follow."

He is right about the precedent. This is the first time either country has formally proposed a bilateral channel specifically for AI-related security events. It follows a May 2026 summit in Beijing where Trump and Xi first discussed the idea of a "Board of Trade" to manage economic friction. Over the weekend, U.S. Trade Representative Jamieson Greer confirmed that both sides had agreed to operationalize that board, suggesting the AI notification mechanism may sit within a broader framework for managing cross-border technology disputes.

But precedent is not the same thing as function. The nuclear hotline worked because both sides understood exactly what they were trying to prevent, and because the cost of getting it wrong was existential. AI incidents are more frequent, more ambiguous, and often more diffuse. A single model weights leak or a jailbreak that enables automated malware generation may not feel like a national security event in the same way a missile launch does, but its downstream effects could be far more widespread and harder to contain.

What Happens Next

The practical details of the notification mechanism have not been released, and may not even exist in draft form yet. Based on how similar systems work, any functional agreement would need to address at least three questions.

First, what triggers an alert? The threshold needs to be low enough to catch genuinely dangerous events, but high enough that neither side is flooding the other with noise about routine model failures or research accidents. Second, who receives the alert? If it goes to diplomatic liaisons, they may lack the technical expertise to evaluate the report. If it goes to technical teams, they may lack the political authority to respond. Third, what happens after the notification? A hotline is only useful if both sides agree on what de-escalation looks like, and right now the United States and China do not even agree on what kinds of AI development should be allowed.

Bessent's proposal is a reasonable first step in an area where doing nothing is increasingly risky. But the comparison to nuclear diplomacy reveals how much harder AI governance will be. Missiles fly in one direction. AI incidents ripple outward in ways no single alert can capture. The hotline may get installed. Whether anyone knows when to pick it up is another question entirely.