US-China AI Safety Hotline Faces Trust Problem Amid Race Rhetoric

Introduction

The United States and China, the two leading powers in artificial intelligence, are trying to preserve a direct line for discussing emerging AI risks. Treasury Secretary Scott Bessent said the countries had discussed a notification mechanism through which either side could alert the other when its AI systems become dangerous or behave unpredictably. The channel could also be used to exchange views on common goals.

That sounds like a modest but useful step toward crisis management. Yet the proposal is emerging amid tighter US technology restrictions, accusations of intellectual-property theft, and repeated political claims that America must “win” the AI race. Those circumstances raise a basic question: can the two governments discuss safety without treating every exchange as part of their strategic contest?

Key points

• The mechanism appears underdeveloped. People familiar with the proposal told reporters that it would not include officials with the technical expertise needed to judge the urgency of an alert. Without specialists, a warning may be transmitted but not properly interpreted.
• China has not publicly embraced the plan. Beijing did not appear to withdraw from talks despite President Donald Trump’s confrontational statements about US dominance in AI. However, it has not publicly acknowledged the proposed notification system, suggesting that political confidence remains limited.
• Technology policy is shaping the safety debate. The US has expanded export controls intended to restrict China’s access to chips and other hardware used for AI development. Congress is considering additional restrictions, while the Justice Department has accused several Chinese AI companies of stealing from US frontier labs—claims China disputes.
• A real system would need shared procedures. Analysts say the countries would have to agree on risk categories, notification thresholds, response protocols, and regular reviews. Otherwise, the arrangement could become a diplomatic gesture rather than an operational safety tool.

Why cooperation still matters

The two countries do not need identical political systems or identical risk perceptions to find areas of mutual interest. Hackers and terrorists using AI to attack infrastructure represent a concern for both governments. The countries may also have complementary experience. Sam Bresnick of Georgetown University’s Center for Security and Emerging Technology noted that China has conducted regular model audits, while the US is only beginning to develop comparable practices for frontier systems.

The obstacle is that Washington and Beijing do not frame AI risk in the same way. US debates often focus on the possibility of uncontrollable or highly capable systems. China places greater emphasis on the state’s ability to supervise and control AI. That difference may narrow as Chinese capabilities advance, but it currently makes a shared safety vocabulary difficult to establish.

Implications

The proposed channel’s immediate value may be limited to keeping communication open during a crisis. It is unlikely to provide reliable assessments of military risks, model failures, or large-scale misuse unless technical experts are directly involved and both sides trust the information process.

Alvin Wang Graylin of Stanford’s Digital Economy Lab argues that continued dialogue could eventually show both countries that AI risk management requires cooperation. Bresnick is more cautious: if China interprets “AI safety” as a code word for constraining Chinese development, Beijing may have little incentive to use the mechanism seriously.

That is the central policy dilemma. The US wants safeguards without slowing its strategic lead, while China is being asked to share safety information as its access to advanced hardware is restricted. A hotline cannot resolve that contradiction by itself. For the proposal to become meaningful, Washington would need to separate technical risk reduction from the broader campaign to limit China’s AI capabilities. Otherwise, the rhetoric of winning the AI race could undermine the safety cooperation it claims to support.

Source: Ars Technica AI