On Monday, OpenAI published a set of proposed international safety standards for artificial intelligence as global leaders — notably the United States and China — consider how to reduce AI-related risks.
The proposal seeks to establish agreed procedures and metrics for classifying, reporting, and responding to incidents that arise during pre-deployment testing and those that occur in real-world operation. According to OpenAI, the standards would cover events occurring both in test environments before deployment and incidents that manifest in operational settings.
OpenAI recommends using existing AI safety institutes around the world to help set those standards, and suggests coordination through the U.S. Center for AI Security and Innovation, which is part of the U.S. Commerce Department.
Diplomatic context and notification mechanisms
The proposal arrives against a backdrop of diplomatic discussions: U.S. Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng reportedly discussed over the weekend a U.S.–China AI dialogue intended to notify the other country when an AI incident reaches national security risk levels. OpenAI’s recommendations aim to support the development of the detailed protocols that such a dialogue would require, by proposing shared metrics and reporting standards.
What OpenAI recommends
- Leverage international AI safety institutes to facilitate standard-setting.
- Use the U.S. Center for AI Security and Innovation as part of the coordination framework.
- Develop shared, global measurements and classification systems for AI incidents.
- Define clear procedures for tracking, reporting, and responding to alignment issues.
An OpenAI official told Axios that the objective is to have international standards covering both incidents that appear in testing before deployment and those that occur in the real world.
Why it matters
Industry recommendations are expected to heavily inform U.S. approaches to coordinating AI risk management with China, so proposals from major AI labs like OpenAI could influence international regulatory and diplomatic responses. Shared metrics and reporting mechanisms could enable faster and more transparent state-to-state communication in the event of serious AI incidents.
What to watch next
Expect further input from AI labs and industry actors in the coming days. All eyes are on meetings involving President Trump and Chinese President Xi Jinping this week, including a state dinner that will feature leading industry figures. Sam Altman is expected to outline OpenAI’s proposals in an address to the United Nations Security Council in New York on Wednesday.
(Consistent with the source: the proposal was released on Monday; Altman’s address is scheduled for Wednesday. The source provided no additional specifics on metrics or exact meeting times.)
