The United States and China are discussing a possible system for notifying each other about serious artificial-intelligence incidents, according to comments by U.S. Treasury Secretary Scott Bessent reported by The Associated Press on September 21.
The proposal emerged as the two countries prepare for high-level diplomatic contacts during the United Nations General Assembly. Washington and Beijing remain strategic competitors in advanced computing, semiconductors, data infrastructure and foundation models, but both governments increasingly acknowledge that highly capable AI systems could create risks that are difficult to manage through national policy alone.
An incident-alert mechanism would likely be narrower than a comprehensive AI treaty. It could focus on events such as the unexpected loss of control over a powerful model, large-scale cyber misuse, dangerous autonomous behavior or other incidents with potential cross-border consequences. The available public reporting did not establish the system’s final design, legal status or whether China had formally agreed to it.
The significance is institutional rather than immediate. Existing strategic hotlines between major powers generally focus on military activity, but advanced AI introduces a new category of risk that can move through commercial cloud infrastructure, open-source software and globally connected networks. A serious incident could spread faster than traditional diplomatic channels are designed to handle.
The initiative also reflects the unusual dual-track character of the U.S.-China relationship. The countries compete to control the most important technologies of the next decade while depending on one another to reduce the danger of miscalculation. Washington has used export controls and investment restrictions to limit China’s access to advanced chips and manufacturing equipment. Beijing has pursued domestic alternatives and tighter oversight of data and algorithms. At the same time, both sides face pressure from companies developing systems whose behavior may not always be predictable.
Any alert arrangement would face difficult questions. Governments would need to define what qualifies as a reportable incident, determine which agencies receive the information and protect commercially sensitive data. Companies would also have to decide how quickly to disclose failures without exposing security vulnerabilities or triggering unnecessary panic.
The proposal is therefore best understood as an early confidence-building measure rather than a solution to the broader AI governance problem. Its value would depend on whether officials can communicate during a real event, when national-security concerns, commercial interests and public pressure are likely to collide.
For the global technology sector, even a limited mechanism could establish a precedent: that AI safety is not only a corporate compliance issue but also a matter of strategic stability between the world’s two largest technology powers.
Sources: - https://apnews.com/article/2c7f54f07e755f506d9db9b91df282bd - https://apnews.com/article/d08e58524de21511dcaba51f43faad95