The United States and China have agreed to establish a dialogue on artificial intelligence, a move aimed at reducing risks between the world’s two leading powers in the technology.
U.S. Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng discussed the proposal during a meeting at JPMorgan Chase’s New York City offices in the lead-up to a September 2026 White House visit by General Secretary Xi Jinping, War on the Rocks reported.
The agreement, detailed in subsequent readouts of the leaders’ meeting, includes the creation of a “U.S.-China Super Intelligence (SI) Dialogue to exchange views on risks and benefits related to SI” and a “bilateral communication channel for SI incidents.” The next exchange is scheduled for November 2026. Both countries affirmed that “AI must be kept under human control,” echoing a 2024 agreement on nuclear weapons.
Confidence-Building in a New Arena
The dialogues do not represent a form of AI arms control. They lack enforcement or verification mechanisms and do not constrain innovation or alter the balance of AI power. Instead, they function as confidence-building measures—voluntary, non-binding steps aimed at reducing opacity and facilitating communication to mitigate the risk of inadvertent escalation.
This approach draws a historical parallel to Cold War-era agreements between the United States and the Soviet Union. In 1963, the two nations established a bilateral hotline for crisis communication, a tool later used to prevent miscalculations during conflicts involving Israel and its Arab neighbors.
This hotline eventually led to the 1987 creation of Nuclear Risk Reduction Centers, a framework later extended in 2013 to include cyber crises. Former President Barack Obama used the hotline in 2016 to warn Russian President Vladimir Putin about election interference.
A Shared Risk, a Rivalrous Race
The establishment of these AI-specific channels comes amid a fierce strategic rivalry. Both countries view AI as central to their national power and military capabilities. AI offers transformative military advantages, including improved intelligence analysis, targeting, and command-and-control functions, as seen in recent conflicts in Ukraine and the Middle East.
Despite this competitive drive, both nations share mutual concerns about AI’s national security risks, particularly its role in nuclear decision-making. An official Chinese position paper from 2021 recognized that AI could deepen strategic miscalculation, undermine mutual trust, and escalate conflicts.
The integration of AI into nuclear command, control, and early warning systems introduces new forms of uncertainty. For instance, using AI tools to support decision-making around nuclear use or provide various courses of action may bias decisions in favor of more escalatory options.
Challenges to a Lasting Framework
While the new dialogues represent progress, their effectiveness faces unique challenges compared to historical confidence-building measures. A primary limitation is the absence of a singular, catalyzing crisis akin to the Cuban Missile Crisis of 1962, which galvanized U.S.-Soviet efforts to institutionalize risk-reduction mechanisms. Without a shared, vivid sense of AI-triggered danger, the depth of mutual understanding on specific risks remains unclear.
Fundamental disagreements persist over key concepts. For example, the United States and China hold different definitions of what constitutes a lethal autonomous weapon. China applies a high threshold, defining unacceptable systems as those that operate across the entire kill chain “with an absence of human intervention and control during the entire process of executing a task.”
The U.S. definition is broader, describing any system that “once activated, can select and engage targets without further intervention by an operator.” These differing conceptual frameworks could hinder the development of common rules of the road.
Another challenge is the voluntary nature of these agreements, which makes them relatively easy to abrogate. However, despite these limitations, the establishment of direct communication channels marks a pragmatic effort to manage risks in a technological domain where the consequences of miscalculation could be catastrophic.
The dialogue’s next exchange is scheduled for November 2026, where the two nations will begin to define the scope and operational details of their bilateral communication on AI incidents.
