The Pulse
OpenAI Urges U.S. to Lead Global Frontier-AI Standards
OpenAI is calling for U.S.-led international standards covering recursive self-improvement, frontier-AI measurement, safeguards and incident reporting.
AI.info Team ·
OpenAI is asking the United States to lead a global system for frontier-AI standards, proposing common methods for measuring model capabilities, assessing safeguards and reporting incidents involving increasingly autonomous systems.
The company’s September 21, 2026, proposal centers on recursive self-improvement, or RSI: a process in which AI systems take on more of the work involved in developing later generations of AI. OpenAI says fully autonomous RSI is not happening today and should not proceed unless people can preserve meaningful control over the research process.
“That is why we believe the United States should lead an effort to work together with countries around the world to develop global technical standards for frontier AI, including for RSI.”
OpenAI, the company’s Global Affairs team
OpenAI does not propose a global regulator or a single approval system. Its plan would create technical standards through cooperation among national AI safety institutes, the Center for AI Standards and Innovation, industry groups, independent experts and academic researchers.
OpenAI’s proposed standards would measure more than model scores
The proposal calls for shared definitions and measurement methods that countries could use to compare frontier systems and evaluate whether their safeguards are sufficient. OpenAI says the standards should cover both open and closed models, and should not favor a specific company, country or business model.
OpenAI identifies three problems that international coordination could address: conflicting national evaluation and reporting rules, incentives for countries to act independently even when the result harms everyone, and uneven access to technical expertise. The company says those problems could become more serious if automated AI research accelerates faster than institutions can assess its effects.
The proposed standards would cover the progress of AI systems on RSI-related tasks, the amount of autonomous research taking place inside an AI company, and the human oversight applied to those processes. They could also define when an automated research activity should trigger immediate human review.
Incident reporting sits alongside capability testing
OpenAI also wants common rules for classifying, tracking and reporting alignment failures and other problems arising from automated AI research. The company points to shared severity levels and reporting thresholds as examples of the information governments and developers would need in order to compare incidents and respond across borders.
The proposal calls for secure communication channels among governments and critical-infrastructure operators to exchange information about national-security concerns, emerging vulnerabilities and threats. OpenAI specifically describes dialogue between the United States and China in these areas as a positive step, saying upcoming talks are taking place at an opportune time.
CAISI would anchor the U.S. effort
OpenAI proposes building on the network of AI safety institutes established in Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India and the United Kingdom. Those institutions could work with CAISI and national industry bodies to develop standards for frontier models and automated AI research.
The company also points to the International Network for Advanced AI Measurement, Evaluation, and Science, created through CAISI in 2024. OpenAI says that existing cooperation could provide a foundation for shared capability benchmarks, risk assessments and judgments about whether safeguards are adequate.
Under the proposal, the resulting standards would not themselves serve as licenses, mandatory prerelease reviews or model-approval requirements. Individual governments would decide whether and how to incorporate them into domestic law, preserving national authority while giving countries a common technical reference point.
OpenAI links global rules to U.S. leadership
OpenAI argues that the United States is well positioned to lead because its AI industry is at the technical frontier and the country occupies a strong position in finance, trade, defense, technology and information systems.
The company says U.S. leadership could determine whether the country helps shape a global AI framework or allows a fragmented and uneven system to develop. It presents strong national governance, connected through practical international cooperation, as a way to support safeguards, continued innovation and broad access to AI’s benefits.
OpenAI says international standards should be developed transparently with input from open-model developers, closed-model developers, independent technical experts and universities. It also recommends cooperation with standards organizations and industry groups including ISO, the Frontier Model Forum, the Agentic AI Foundation, the Open Secure AI Alliance and the Appia Foundation.
The unresolved question is adoption
OpenAI’s plan offers a framework for technical cooperation but does not resolve how governments would enforce the standards, which systems would qualify as frontier models or what evidence would trigger a reporting obligation. The company says capability benchmarks could help define which developers and models fall within scope, but it does not set a threshold in the announcement.
The proposal also leaves national governments to determine whether common standards remain voluntary or become part of binding regulation. That distinction will decide whether the system functions mainly as a shared measurement language or as the foundation for formal oversight.
For now, OpenAI is asking Washington to organize the process and use U.S. institutions to connect national safety bodies across borders. The company’s immediate proposal is a standards network, common incident protocols and a U.S.-led diplomatic effort—not a global licensing regime.