OpenAI Urges US to Lead Global Standards for Advanced AI Safety

OpenAI has called on the United States to spearhead an international effort to create global technical standards for frontier artificial intelligence systems. The proposal, published Monday, coincides with the UN General Assembly and the start of diplomatic talks between US and Chinese officials on AI safety coordination. The standards would focus on capability measurement, risk assessment, and safeguard sufficiency for advanced models.

A central element of the proposal addresses recursive self-improvement (RSI), where AI systems automate the development of subsequent AI generations. OpenAI states that fully autonomous RSI does not currently exist and should not be pursued until it can be done safely. The company warns that without adequate oversight, humans could lose practical control over AI development and fail to understand the research processes conducted by the systems themselves.

“Without adequate oversight, humans could lose practical control over AI development and cease to understand the research processes conducted by the systems.”

The framework envisions complementary national and international standards, common metrics, and incident communication protocols. OpenAI emphasizes these technical standards would not function as licenses or mandatory pre-release approvals; national governments would decide how to incorporate them into legal systems. The company suggests building upon the International Network for Advanced AI Measurement, Evaluation and Science, which includes the US Center for AI Standards and Innovation (CAISI) and organizations from nine other governments.

OpenAI argues the US is well-positioned to lead due to its technical frontier status and privileged position in finance, trade, defense, and technology. The company notes that geopolitical competition will increasingly hinge on AI adoption and diffusion, driven by realistic cooperation and sensible rules.