U.S. and China propose establishing a bilateral hotline and 'AI red lines' to de-escalate AI-related geopolitical tensions and manage competition during formal diplomatic summit.

The U.S. and China have proposed developing a hotline for AI incident reports to address the risk of rogue AI models escaping oversight and to prevent misinterpretations of autonomous actions that could trigger rapid escalation without human intervention. The proposal comes ahead of Chinese Premier Xi Jinping's visit to Washington on September 24, with U.S. Treasury Secretary Scott Bessent confirming that follow-up AI safety discussions are scheduled between the two nations within two months.

The proposal emerged from discussions between Bessent and Chinese Vice Premier He Lifeng in New York. The two sides have proposed that President Trump and Xi Jinping discuss various AI safety mechanisms, including a notification system focused on national security. Such a system could include warnings of detected rogue AI actors operating anywhere in the world, as well as notifications if an autonomous system performed an action that could be misinterpreted as an attack.

Both nations seek to prevent a scenario in which an AI-driven autonomous weapon system—whether kinetic or asymmetric in nature, such as AI-driven cyber attack capabilities—responds to provocation without human consideration. Such an action could trigger an escalated response from another AI system, leading to rapid escalation to full-blown conflict at a pace where human intervention would be impossible.

The proposed system could also provide a framework for joint responses to potential threats and positive goals, potentially forming the basis for coordinated development pacing. However, China has expressed reservations about calls from U.S. AI firms to slow AI development, suggesting such measures could be used to hamper Chinese capabilities. While the extent of China's commitment to the notification system remains unclear, analysts and experts on both sides have advocated for such a framework.

Early discussions have produced only the bare bones of a proposal. Much remains to be negotiated on alignment between the two major AI-developing nations ahead of the summit. Some analysts remain skeptical of firm agreements given limited trust between the parties, though there is hope for consensus on banning direct weaponization of AI, establishing core safeguarding principles, and preventing autonomous systems from targeting critical infrastructure. Representative Ro Khanna of the House Democratic caucus called on the two leaders to develop core AI principles, specifically banning recursive self-improving AI and preventing its use on biological and nuclear weapons technology. The bipartisan interest in the outcome suggests that finding common ground on AI governance could set the stage for broader discussions on economic relations between the two nations.