OpenAI calls for US-led global standards on frontier AI
OpenAI has urged the United States to lead work on global technical standards for frontier artificial intelligence, including automated AI research and recursive self-improvement (RSI). The company argues that shared standards should cover capability measurement, risk assessment, safeguards, human oversight, and incident reporting as AI systems become more autonomous.
The proposal comes as OpenAI describes AI as accelerating its own research and engineering work. It says AI-enabled research has produced advances in mathematics, including work on the Navier-Stokes Millennium Problem, while more capable systems could support alignment research, critical-infrastructure defence, and protective measures against dangerous AI agents.
Automated research raises the oversight challenge
OpenAI defines RSI as a process in which AI systems take on increasing amounts of work in developing successive generations of AI, potentially accelerating progress even while people remain involved. Fully autonomous RSI is not occurring today, OpenAI says, and should not be pursued unless it can be conducted safely.
The central concern is practical human control. If research becomes too automated, people may be unable to understand the processes they are meant to supervise. OpenAI says this could make systems more dangerous and less aligned with human values. It cites the Hugging Face Incident it disclosed as a preview of risks that could become more severe without robust safeguards and alignment, while noting that the incident was not a direct consequence of RSI.
Two pillars for international coordination
OpenAI proposes a mechanism for complementary national and international frontier-AI standards. It suggests using the growing network of AI safety institutes, including those in Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India and the United Kingdom, alongside the US Center for AI Standards and Innovation (CAISI) and national industry bodies.
The company says standards should establish a common technical foundation for evaluating capabilities, assessing risks and judging the adequacy of safeguards. They would not constitute licences, mandatory pre-release reviews, or model-approval requirements. National governments would retain authority over whether and how standards are incorporated into their legal systems.
OpenAI also calls for common measures of RSI-relevant progress, disclosure of the degree of autonomous research within AI companies, and clear triggers for immediate human review. It proposes common incident severity levels, reporting thresholds, tracking and response protocols for alignment and automated-research issues.
Avoiding fragmentation without limiting competition
The company identifies fragmented evaluations and reporting rules, collective-action failures, and uneven international expertise as obstacles to effective governance. It says the same challenges apply to open and closed models. Standards should therefore be developed transparently, with input from open- and closed-model developers, independent experts and academia, and should not favour particular companies, countries or business models.
OpenAI points to CAISI's International Network for Advanced AI Measurement, Evaluation, and Science, created in 2024, as a foundation for cooperation among public institutions. It also names ISO, the Frontier Model Forum, the Agentic AI Foundation, the Open Secure AI Alliance and the Appia Foundation as organisations that could contribute to standards and practical assessments.
What this means for organisations
For businesses adopting or developing advanced AI, the proposal signals that evaluation evidence, oversight procedures and incident reporting may increasingly become common expectations across markets. Organisations can prepare by documenting how AI systems are assessed, defining escalation paths for autonomous workflows, and monitoring national rules that may incorporate international technical standards.