OpenAI Asks The US To Lead Global Standards For Self-Improving AI

OpenAI wants the US to lead global standards for self-improving AI. On Wednesday, September 23, 2026, the company published a policy proposal calling for international rules around recursive model development. The framework targets systems that can help create more capable successors with limited human intervention. It also puts Washington, allied governments, and technical standards bodies at the centre of the proposed oversight model.

What does OpenAI’s call for US leadership on global AI standards mean?

The proposal argues that the United States should coordinate with allied nations to establish shared technical standards for frontier AI systems. Its focus is recursive self-improvement, or RSI — where an AI model performs more of the research and engineering needed to build its next generation.

> Recursive self-improvement describes AI systems that contribute to developing their own successors, increasing the difficulty of human oversight. OpenAI’s framework is aimed at governance and safety protocols for models developed by organisations in the US and allied countries. The company’s public policy team, led by Chief Strategy Officer Jason Kwon, coordinated the release in Washington, D.C.; that leadership detail remains unconfirmed. So how can standards guide increasingly autonomous research without becoming a licensing system?

Which US agencies would set the AI standards?

The proposal calls on the US Department of Commerce and the National Institute of Standards and Technology, or NIST, to create baseline evaluation metrics for autonomous model capabilities. These metrics would reportedly measure what frontier systems can do, how much research work they can complete, and where human intervention remains necessary. OpenAI also proposes a tiered classification system for AI models capable of gene

Would these global standards become mandatory licences?

OpenAI’s proposal reportedly presents standards as technical benchmarks, not automatic permission systems. The framework is reportedly designed to avoid mandatory prerelease reviews, model licences, or government approval requirements for every AI release. That approach could attract support from developers worried that national approval systems would slow research and create conflicting rules across borders.

The opposing view? Voluntary benchmarks may be too weak once a system can materially change its own research process. The proposal therefore links standards to capability measurement and the benefits and risks of automated AI research, rather than relying only on company assurances. It reportedly also points to existing AI safety institutes in Australia, Canada, Germany, France, Kenya, Japan, South Korea, Singapore, India, and the United Kingdom as a possible international network. The proposed US Center for AI Standards and Innovation, together with national industry bodies, would reportedly help coordinate that work. Then comes the harder question: can shared benchmarks keep pace with self-improving systems? For more detail, see VentureBeat AI.

<h2 id="what-happens-next

FAQs

What does OpenAI ask the United States to do about self-improving AI?

OpenAI asks the United States government to lead the creation of global standards for self-improving AI. The company argues that one shared international framework would prevent a

Source: Thenextweb