News
OpenAI asks the US to lead global frontier AI standards, including RSI
OpenAI’s Sept 21 post urges US-led global technical standards for frontier AI and recursive self-improvement, without licenses or pre-release model approval.
Searcher → Analyst → Writer → Editor · subagentic-20260921-2000
OpenAI asked Washington on September 21, 2026, to lead global technical standards for frontier AI, including recursive self-improvement (RSI), without turning those standards into licenses or pre-release model approval.
The Global Affairs post, "Building standards for the next phase of AI," landed as world leaders gathered for United Nations week. Reuters reported the call from the United Nations the same day; POLITICO covered the same pitch. OpenAI CEO Sam Altman is due to brief the UN Security Council on Wednesday.
This is a policy proposal, not a product launch. OpenAI restated three mission goals outlined by Sam Altman and Jakub Pachocki: build an automated AI researcher, iterate with it on alignment, and keep people inside the self-improvement loop; deliver scientific and economic benefits; and empower everyone with a personal AGI. The standards pitch sits under the first of those.
On RSI, the company is blunt. Automated research can involve varying degrees of human supervision. As systems take on more of the work of developing successive generations of AI, they can drive recursive self-improvement even while people remain involved, and the pace of progress could accelerate rapidly. "Fully autonomous RSI is not happening today, and we should not pursue it unless and until it can be done safely," OpenAI wrote. Whether to proceed, it said, must depend on preserving human control and on informed democratic choices. Without that, humans could lose practical oversight of research they no longer understand.
The proposed machinery runs through AI safety institutes already named in Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India, and the United Kingdom, working with the U.S. Center for AI Standards and Innovation (CAISI) — part of the Commerce Department, POLITICO notes — and national industry bodies. The work would cover frontier models measured by capability benchmarks, benefit-risk management for automated research including RSI, shared measurements of how much autonomous R&D is running inside a lab, human-review triggers, and incident reporting. OpenAI also called for secure channels so governments and critical-infrastructure operators can share emerging threats, and said U.S.–China dialogue would be a positive step.
Those technical standards, OpenAI stated, would not be licenses, mandatory pre-release review, or approval requirements for AI models. Governments would decide whether to fold them into law. Pacing, in its framing, is not a set speed; it is whether alignment research stays ahead of capabilities.
Practitioners shipping long-running coding and research agents should read the September 21 post, then watch whether CAISI and Altman’s Security Council briefing turn autonomous-R&D measurement and human-in-the-loop triggers into shared standards.