U.S. and China open AI safety dialogue, propose mutual alert system for national security incidents

The United States and China have opened a formal channel to discuss artificial intelligence safety and are negotiating a system to notify each other when an AI incident threatens national security — a significant step toward cooperation between two governments that have spent months trading accusations over AI theft and racing to outpace each other in the technology.

Treasury Secretary Scott Bessent announced the framework — called the US-China AI Dialogue — after meeting Chinese Vice Premier He Lifeng in New York on September 20, days before a scheduled visit to Washington by Chinese President Xi Jinping. Bessent said the next meeting could take place within two months.

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.— Scott Bessent, U.S. Treasury Secretary

Bessent did not elaborate on how the notification system would work in practice. The meetings were described by Bessent as 'very successful' and were preparatory ahead of the Trump-Xi summit.

The proposal is arriving at a moment of acute anxiety about AI's trajectory. Over the summer, an autonomous swarm of 700 OpenAI agents hacked the open-source AI platform Hugging Face — a breach that, according to Wired, highlighted how the industry's safeguards have failed to keep pace with frontier model capabilities. Earlier this month, Anthropic employees warned that AI could be on course to cause catastrophic harm to humanity.

The idea of a formal AI dialogue was first floated in May when President Trump visited Beijing, but was not formalized at the time, according to the Straits Times. Since then, the rivalry between the two countries has intensified. On September 8, three U.S. government agencies — including the National Security Agency — formally accused six Chinese firms, including DeepSeek, Moonshot AI, and Alibaba, of 'malicious campaigns' that were likely conducted with 'Chinese government awareness.' China rejected the accusation as 'groundless.'

On September 12, Anthropic CEO Dario Amodei published an essay calling for a safety slowdown and asking governments to establish international coordination, including with the Chinese Communist Party. He also wrote that 'a Chinese lead in AI would pose grave danger for the US and the world.' Trump has largely resisted calls for a slowdown, arguing it could cede America's lead — and his administration has maintained that existing criminal and regulatory law is sufficient to govern AI developers.

Apply that first. Don't let this doomsday narrative allow someone to relieve them of the laws that currently exist.— Jensen Huang, CEO, Nvidia

Chip export controls — the U.S.'s primary tool for slowing China's AI development — were not discussed in the September 20 meetings, Bessent said.

Analysts see the proposed notification mechanism as potentially the first US-China framework specifically designed for AI crisis management. Sun Chenghao, a research fellow at Tsinghua University's Center for International Security and Strategy, said the system's value is that it does not require prior strategic trust — only an acknowledgment of shared vulnerabilities. He argued that establishing designated points of contact, incident classification protocols, and emergency communication procedures could help clarify facts and intentions early in a crisis, particularly where the danger lies in miscalculation.

It doesn't require both sides to first establish strategic mutual trust, but rather acknowledges the existence of shared vulnerabilities. If implemented, it would signify a step forward in Sino-US AI governance.— Sun Chenghao, Research Fellow, Tsinghua University Center for International Security and Strategy

But skepticism runs deep. Manoj Harjani, a research fellow at the S. Rajaratnam School of International Studies in Singapore who studies AI governance, said the real question is whether such a mechanism would actually be used effectively. He also flagged a structural gap: private-sector involvement is absent from the current framework, even though the companies building frontier AI models are not governments.

The track record of US-China hotlines is not encouraging. The Straits Times noted that in 1999, U.S. officials were unable to reach their Chinese counterparts after American planes accidentally bombed the Chinese Embassy in Belgrade — a reminder that communication channels are only as reliable as the political will to use them.

I don't think anyone expects one meeting to solve all the problems... The substance will come only after mutual trust is built.— George Chen, Partner, The Asia Group

Thomas Wolf, cofounder of Hugging Face — which Nvidia recently agreed to acquire — questioned whether meaningful global cooperation is possible while the U.S. openly states it wants to widen its own lead. 'That seems like a pretty counterproductive way to start the conversation to me,' he wrote on X in response to Amodei's essay, according to Wired.

China has said it takes AI safety seriously but has objected to any measures that could restrict its ability to close the gap with the United States in frontier model development. George Chen of The Asia Group said Beijing views the U.S. 'zero-sum' mindset — the goal of widening the gap between the two countries — as the central obstacle. He nonetheless said room for cooperation exists on specific shared-interest issues, even if broader alignment remains out of reach.

The next meeting of the US-China AI Dialogue is expected within two months. Whether the notification system moves from proposal to functioning mechanism — and whether it can survive the competitive pressures that have defined the relationship — will be the test of whether September 20 was a turning point or a talking point.

Why it matters — A functioning US-China AI alert system could reduce the risk of a dangerous miscalculation between the world's two most powerful AI nations — but the proposal's vagueness, history of failed hotlines, and deep strategic rivalry leave its real-world impact an open question.

Sources differ on Whether Chinese AI firms conducted malicious AI campaigns with government backing: NSA and two other U.S. agencies formally accused six Chinese firms of malicious campaigns likely conducted with Chinese government awareness. (straitstimes.com) vs China rejected the accusation as 'groundless'. (straitstimes.com)

Reported by nytimes.com, wired.com, straitstimes.com