The United States and China are moving towards a formal dialogue intended to reduce the risk of dangerous artificial-intelligence incidents escalating into an international crisis.

US Treasury Secretary Scott Bessent said American officials proposed a notification mechanism during talks with Chinese Vice Premier He Lifeng in New York.

Under the proposal, the two governments would communicate when an AI related incident reached a level that could affect national security.

Senior officials are expected to meet again in Shenzhen, China, in approximately two months to discuss AI dangers and possible communication protocols.

The initiative is being developed ahead of a meeting between President Donald Trump and Chinese President Xi Jinping.

The announcement does not mean the countries have agreed on final definitions, enforcement rules or reporting deadlines.

Officials would still need to determine what qualifies as a reportable incident, which agencies would communicate and how confidential technical information would be protected.

Potential concerns include autonomous systems behaving unpredictably, AI assisted cyberattacks and malicious use by groups operating outside government control.

A communication channel could serve a purpose similar to crisis hotlines used in other areas of international security. Its value would be to reduce uncertainty when a dangerous event occurs and help both sides distinguish an accident, criminal act or technical failure from deliberate government action.

The United States and China remain competitors in advanced chips, computing infrastructure, robotics and AI models.

Both governments consider technological leadership an economic and national security priority. That rivalry may make them reluctant to share information about weaknesses in important systems.

Bessent said greater transparency between the world’s leading AI powers was important. He also argued that developers should remain responsible for the behaviour and security of the systems they produce.

An incident line would not replace domestic safety rules, independent testing or corporate accountability. It would provide a government communication route after a serious event had occurred or been discovered.

The plan could become one of the first continuing arrangements specifically addressing AI related security risks between Washington and Beijing.

Its success will depend on whether both governments provide accurate information when doing so may reveal a technical vulnerability or politically embarrassing failure.

For now, the proposal is an early diplomatic framework rather than a binding international safety agreement.