The United States and China are preparing to advance discussions on superintelligence at next week's summit in Shenzhen, seeking common approaches to managing risks from increasingly capable artificial intelligence systems. The initiative matters because the two competing powers are exploring safety cooperation while continuing to compete for technological leadership.
U.S. Treasury Secretary Scott Bessent announced the planned progress during a speech to the National Committee on U.S.-China Relations in New York, according to Chosunbiz, citing Reuters. The discussions will take place alongside the November 18–19 Asia-Pacific Economic Cooperation (APEC) leaders' meeting.
What the talks cover
The proposed dialogue will focus on practical safeguards rather than aligning every aspect of American and Chinese AI development. Bessent said both governments want to reduce shared risks, avoid misunderstandings between their systems and establish common expectations for evaluating advanced AI models.
A central priority will be agreeing on best practices for assessing and testing advanced AI before public release. The discussions will also address how to preserve human oversight, ensuring people retain decision-making authority over the technology.
A channel for emergencies
The two governments have also agreed to establish a direct communication mechanism for major AI incidents. It would allow either side to alert the other quickly when a significant event occurs, giving the countries a way to communicate about potential problems without waiting for broader diplomatic discussions.
The arrangement builds on an understanding reached by the American and Chinese leaders during a September White House summit, when they agreed to create a new AI alert system.
However, cooperation has not eliminated Washington's concerns about maintaining its technological advantage. At the September meeting, President Donald Trump expressed caution about introducing additional safeguards that could affect American leadership in AI.
Competition remains a barrier
Bessent presented the dialogue as an attempt to balance technological competition with international responsibilities. He argued that increasingly capable systems will create new categories of risk that require governments to develop ways of managing them.
He also acknowledged that the two countries do not share identical interests or approaches to AI development. That distinction matters: discussions about testing standards and incident notifications do not necessarily mean agreement on broader rules governing the technology.
Why it matters
The Shenzhen meeting could move U.S.-China AI diplomacy from general commitments toward practical cooperation on model evaluations, pre-release testing and emergency communications. These areas are particularly important as both countries develop increasingly advanced systems.
The next milestone is the November 18–19 APEC summit, where the bilateral dialogue is expected to advance. Its progress will show whether Washington and Beijing can establish shared safety practices while preserving their competing ambitions in artificial intelligence.