OpenAI has proposed international standards for artificial intelligence safety amid discussions between the United States and China on mechanisms for jointly responding to risks associated with AI development. The company believes countries need common approaches to classifying incidents, tracking them, and sharing information about model safety issues.

OpenAI’s proposals came after talks between U.S. Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng. The parties discussed the possibility of creating a communication channel through which the countries could notify each other about AI incidents that reach the level of a national security threat, Axios reports.

Common Rules for AI Incidents

OpenAI proposes involving AI safety institutes operating in different countries to develop shared standards. According to the company, this process could be coordinated by the U.S. Center for AI Safety and Innovation under the Department of Commerce.

This is not only about formal rules. OpenAI proposes developing common international metrics that could be used to classify AI incidents and determine their severity.

Companies and government agencies should also use comparable approaches to tracking, documenting, and addressing problems related to model behavior that does not align with intended goals.

According to an OpenAI spokesperson, international standards should cover both issues identified before a system’s release—for example, during testing—and real-world incidents after deployment.

The U.S. and China Discuss Direct Information Sharing

OpenAI’s proposals come at a time when Washington and Beijing are considering closer cooperation on AI safety.

During weekend talks, Bessent and He Lifeng discussed the idea of a U.S.-China dialogue on artificial intelligence risks. One of the mechanisms under consideration would have the countries notify each other if an AI-related incident reaches the level of a national security threat.

At the same time, the specific parameters of such interaction remain under negotiation. This is where, Axios notes, the details matter most: it is necessary to determine which events should be considered critical, what information should be disclosed, and how countries should respond to such cases.

The AI Industry Prepares to Put Forward Its Own Rules

OpenAI expects to present its proposals at the international level. The company’s CEO Sam Altman is expected to speak about them on Wednesday at a United Nations Security Council meeting in New York.

Other AI developers are also expected to voice their positions more actively in the coming days. The industry’s particular focus is tied to the upcoming meetings between U.S. President Donald Trump and Chinese President Xi Jinping, as well as to a state dinner that will be attended by the leaders of the largest technology companies.

Thus, the issue of international rules for AI is gradually moving beyond a discussion among individual technology companies. The United States and China are simultaneously discussing communication channels for potentially dangerous incidents, while AI developers are trying to shape standards that could become the foundation of a future international approach to the technology’s safety.