The United States has proposed creating a bilateral notification mechanism through which Washington and Beijing could alert each other about artificial-intelligence incidents that reach the level of a national-security threat. The proposal emerged from talks in New York between US Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng on September 20, ahead of an expected meeting between President Donald Trump and Chinese President Xi Jinping in Washington on September 24.
Bessent said the mechanism was intended to introduce greater transparency between the world’s two largest AI powers and establish a shared understanding of common threats. The proposal remains preliminary: neither government has announced an agreement, operating protocol, reporting threshold or timetable. China’s public account acknowledged that AI was discussed but did not indicate whether Beijing had accepted the proposed mechanism.
A functioning arrangement could cover incidents such as AI-enabled attacks on critical infrastructure, the misuse of advanced systems for biological threats, serious failures involving frontier models, or events in which operators lose effective control of an AI system. Finance, electricity, telecommunications and transportation are among the infrastructure sectors identified by Chinese governance documents as potentially exposed to frontier-AI risks.
The talks represent a material development from the preparatory discussions reported earlier. US officials have now placed a specific confidence-building measure on the table, although fundamental divisions over technology competition remain unresolved. Washington continues to restrict China’s access to advanced processors and semiconductor-manufacturing equipment, while Beijing has accelerated the development of domestic AI models, chips and computing systems. US officials and technology companies have also accused Chinese developers of extracting capabilities from American models through distillation, allegations rejected by China.
The incident-notification proposal does not address those commercial and strategic disputes. Its narrower purpose would be to maintain communication when an AI event could produce cross-border or national-security consequences. Officials have not disclosed whether alerts would cover privately developed systems, military applications, cyber operations, autonomous agents or only the most severe frontier-model incidents. Verification, confidentiality and the treatment of commercially sensitive information would also require negotiation.
The proposal enters a global policy environment in which India has advocated multilateral cooperation on trustworthy AI. The declaration adopted at the India AI Impact Summit in February was endorsed by 92 countries and international organisations, with trustworthiness and international engagement among its central pillars. The September BRICS New Delhi Declaration similarly called for cooperation on safe, secure, inclusive and reliable AI systems.
