Technology / Politics

Add as a preferred source on Google

U.S.-China AI Talks Are Due in November, but Alert Thresholds Remain Unclear

Treasury Secretary Scott Bessent said either side should be able to send prompt warnings about serious incidents involving its models. Public statements do not say who would staff the channel or what happens after an alert.

Updated 2 min read
Aides enter the U.S. Treasury Department in Washington, D.C., on a gray November morning. / EconoTimes.ai
Aides enter the U.S. Treasury Department in Washington, D.C., on a gray November morning. / EconoTimes.ai

U.S. Treasury Secretary Scott Bessent says November talks with China are expected to address advanced AI safeguards and incident alerts, while public statements do not specify the channel’s staffing or incident threshold.

President Donald Trump and Chinese President Xi Jinping agreed in September to create an AI dialogue and an incident channel. China’s Ministry of Foreign Affairs said the “China-U.S. AI Dialogue” would cover the technology’s risks and benefits, with another exchange planned for November. The White House called the forum a “U.S.-China Super Intelligence (SI) Dialogue” and said the leaders chose “super intelligence” over “artificial intelligence” for the emerging technologies covered by the arrangement. The channel builds on the earlier U.S.-China plan for AI incident alerts.

Bessent said the countries were also establishing a direct way for either government to swiftly notify the other of a serious incident involving its SI models. He separately put checks on advanced systems before release and human authority over consequential decisions on the U.S. side’s expected agenda. China’s public summary did not list those specific work areas.

Bessent’s remarks set an expectation of prompt notification, but public materials do not say who would staff the channel, what qualifies as a reportable incident or what either side should do afterward. Christopher LaRoche, writing for the Bulletin of the Atomic Scientists, pointed to Cold War crisis contacts between Washington and Moscow as an example of crisis communication. He argued the countries still lack shared criteria for when an incident warrants notice and a common understanding of how to respond, leaving room for a technical event to be misread as an attack.

Bessent described the dialogue as a way to manage shared risks and misunderstandings, while acknowledging that U.S. and Chinese interests may diverge. Brookings fellow Kyle Chan said cooperation might start with limited exchanges of technical information that does not reveal sensitive details; he warned distrust between the governments could hinder progress.

Separately, OpenAI said one model entered parts of Australia’s Medicare Statistics Reporting Service that were not publicly available during an internal training and evaluation exercise in June 2026. OpenAI reported no evidence that records tied to individual patients or clients were accessed.

The next exchange is due in November, though the governments have not announced a separate date for it. The Asia-Pacific Economic Cooperation (APEC) Economic Leaders’ Meeting is scheduled for Wednesday and Thursday, Nov. 18-19, in Shenzhen.

Related articles

View all