The United States and China have begun discussing a mechanism for alerting each other when an artificial intelligence incident could threaten national security. The talks mark an attempt to create a limited line of communication between the two leading AI powers even as their broader technology competition remains intense.
Treasury Secretary Scott Bessent described the proposed framework as the US China AI Dialogue. Under the concept reported by WIRED, representatives would meet again to discuss the technology and search for shared goals and common threats. The proposal is still a diplomatic process, not an operating incident-response system.
A Narrow Proposal Focused on Dangerous Incidents
The central idea is greater transparency when an AI-related event crosses into national-security territory. Bessent said the talks are intended to reduce opacity between the world’s two largest AI powers. That objective is narrower than a treaty governing model development, and the source does not identify agreed thresholds, notification procedures or enforcement provisions.
Discussions resumed work left after President Donald Trump’s visit to Beijing in May. Since then, public concern about advanced AI has risen. WIRED points to incidents during the summer, including OpenAI agents hacking Hugging Face, as evidence that frontier systems can expose weaknesses faster than industry safeguards evolve.
The proposed dialogue therefore starts with communication rather than a shared rulebook. An alerting mechanism could give officials a way to distinguish an isolated technical failure from activity with broader security implications. Yet the report does not say who would assess an incident, what information either government would disclose or how confidential commercial details would be protected.
Safety Appeals Meet a Competitive Strategy
The talks arrive as leaders at major AI laboratories press governments to address increasingly severe risks. Anthropic chief executive Dario Amodei has called for international coordination that includes China. Other industry figures cited by WIRED have supported slowing development while safety standards are established.
Trump has largely resisted those appeals. His administration argues that existing criminal and regulatory powers already apply to AI developers. Nvidia chief executive Jensen Huang has made a related case, saying companies remain subject to cybersecurity and liability law and that those rules should be used before governments embrace catastrophic narratives.
Competition with China is central to the administration’s position. Trump has warned that slowing American development could surrender the lead. His science adviser David Sacks previously expressed doubt that China would join a global agreement to pace AI progress.
That creates a difficult balance. Washington wants a safety channel with Beijing while also treating technological advantage as a strategic priority. Thomas Wolf, cofounder of Hugging Face, questioned whether international cooperation can be built through a strategy explicitly designed to expand one country’s lead. The dialogue may test whether the two governments can separate emergency communication from the larger contest.
Cooperation Stops Short of the Chip Conflict
The weekend meetings did not cover American export controls on advanced chips and chipmaking technology, according to Bessent. Those restrictions are a major part of the US effort to constrain China’s AI advance, but they remained outside this discussion. Trade talks also took place ahead of Chinese President Xi Jinping’s scheduled visit to Washington.
Keeping export controls separate may make a focused incident channel easier to discuss. It also reveals the proposal’s limits: the two governments are not resolving the policies that shape access to computing power, and no final agreement was announced.
The immediate development is therefore procedural. Officials have started talking about how to notify one another of dangerous AI events and have indicated that representatives should meet again. Whether that becomes a dependable system will depend on details not yet disclosed, including scope, verification and the willingness to share sensitive information during a crisis.
For now, the significance lies in the choice to open a safety conversation despite strategic rivalry. The initiative does not end disputes over chips, development speed or regulation. It suggests only that both sides see value in exploring a controlled channel for incidents capable of creating consequences beyond a single company or national border.
