By CoinAINews Editorial Team
OpenAI is calling for greater international coordination on artificial intelligence safety as frontier models become more capable and AI systems take on a larger role in research and development.
In a policy proposal published on September 21, 2026, OpenAI urged the United States to help lead an international effort to develop technical standards for frontier AI, including systems that could eventually contribute to recursive self-improvement (RSI).
The proposal focuses on common ways to measure AI capabilities, evaluate risks, monitor increasingly automated AI research and report serious incidents. OpenAI argues that compatible standards could help governments, companies and independent organizations evaluate increasingly capable systems across national borders.
The timing is notable. The proposal comes during the United Nations General Assembly's High-Level Week, scheduled for September 22–29, 2026, in New York. The General Assembly's general debate is scheduled for September 22–26 and September 28. United Nations General Assembly
What Did OpenAI Propose?
OpenAI is not proposing a single global AI regulator or a worldwide AI law.
Instead, the company is calling for countries and AI developers to work toward compatible technical standards that can be used across jurisdictions.
OpenAI's September 21 proposal calls for common measurements of frontier AI progress, approaches for evaluating risks, methods for preserving human control and mechanisms for reporting significant incidents.
The company also wants standards that can help determine when advanced AI development should slow down or stop if safety requirements cannot be met.
Reuters reported on September 21, 2026, that OpenAI was calling for the United States to take a leading role in developing global technical standards for frontier AI, including recursive self-improvement.
Why Recursive Self-Improvement Is Getting Attention
One of the most important concepts in OpenAI's proposal is recursive self-improvement, or RSI.
In broad terms, recursive self-improvement describes a future scenario in which AI systems can contribute to improving their own capabilities or help develop increasingly capable successors, potentially reducing the amount of direct human involvement required in AI research.
That does not mean today's AI systems are independently redesigning themselves without human oversight.
OpenAI explicitly states that fully autonomous recursive self-improvement is not happening today and says it should not be pursued unless and until it can be done safely.
At the same time, OpenAI says AI is already accelerating parts of the research used to develop and align future models. The company has described this as evidence of the direction in which AI research could move, while distinguishing it from fully autonomous RSI.
OpenAI Wants Common Measurements for Frontier AI
A major challenge in international AI governance is that different organizations and governments can use different definitions and measurements when evaluating advanced systems.
OpenAI's proposal calls for more consistent technical measurements that could help policymakers understand how quickly AI capabilities are advancing.
For recursive self-improvement, the company has discussed measuring progress related to autonomous AI research and the amount of research work being performed by AI systems within AI-development organizations.
Such measurements could eventually help distinguish between an AI model that primarily assists human researchers and a system performing increasingly important portions of the research process itself.
Human Oversight Would Remain Important
OpenAI's proposal also emphasizes human oversight of increasingly automated AI research.
As AI systems perform longer sequences of tasks, it can become more difficult for researchers to manually inspect every experiment, decision or intermediate result.
Common technical standards could eventually help define situations in which automated AI research should trigger additional human review.
The proposal does not itself create a legally binding international review system. Instead, OpenAI is advocating technical standards that governments could consider when developing their own policies.
AI Incident Reporting Could Become More Standardized
Another important part of the discussion is incident reporting.
OpenAI wants common approaches for identifying and reporting significant AI safety and alignment incidents, including clearer definitions of severity and reporting thresholds.
The basic idea is that if an advanced AI system produces a serious or unexpected behavior, researchers and policymakers should have a shared way to describe what happened and determine its significance.
OpenAI separately published a model misalignment reporting framework on September 16, 2026. The framework describes how the company intends to track, investigate and disclose concerning model behavior, and OpenAI published six initial examples alongside it.
The company said the framework is intended to contribute to broader understanding of AI alignment and potentially help establish more consistent disclosure practices across the industry. OpenAI's September 16 framework
What Does AI Alignment Mean?
AI alignment generally refers to efforts to make AI systems behave consistently with intended goals, safety requirements and human-defined constraints.
The issue becomes more complicated as models become more capable and autonomous.
An AI system that answers questions gives humans relatively direct control over its actions. An AI system capable of conducting extended research, writing and testing code, running experiments and adapting its approach could require different monitoring methods.
OpenAI's policy push is therefore connected to a broader question: how should society measure and manage AI systems when the systems themselves increasingly participate in the process of developing future AI?
Chris Lehane's September 9 Policy Push
The September 21 proposal was not an isolated policy announcement.
On September 9, 2026, OpenAI Chief Global Affairs Officer Chris Lehane published a policy post outlining the company's broader approach to AI governance.
Lehane wrote that OpenAI wanted to advance industry-led frontier AI standards while also advocating compatible international approaches to measuring capabilities, managing risk, preserving human control and determining when development should slow or stop.
He also argued that industry-led standards should complement, rather than replace, mandatory government safeguards and democratic oversight.
That September 9 position provides important context for the September 21 international standards proposal: OpenAI is pursuing several layers of AI governance rather than relying exclusively on voluntary industry commitments.
Why OpenAI Wants International Cooperation
AI development is now a global industry.
AI models can be developed in one country, trained using infrastructure spanning several regions and deployed to users around the world. Researchers, computing resources and technical expertise also move across national borders.
OpenAI argues that fragmented technical standards could make advanced AI safety evaluations harder because companies and governments may use incompatible definitions, tests and reporting systems.
The company therefore wants compatible national and international approaches for measuring capabilities, managing risks, maintaining human control and determining when development should slow down.
OpenAI had already been working on shared AI standards before the September proposal. In June, the company described efforts involving the Appia Foundation and other standards organizations aimed at making AI assessment practices more interoperable across organizations and jurisdictions.
OpenAI — Helping Build Shared Standards for Advanced AI
Why the United States Is Central to OpenAI's Proposal
OpenAI specifically wants the United States to take a leading role in the standards effort.
The company argues that the United States has significant influence across technology, research, finance and international alliances, giving it an opportunity to shape how frontier AI standards develop.
Reuters reported on September 21 that the proposal was released as world leaders gathered at the United Nations, with AI expected to be an important subject of international discussions.
Reuters also reported that OpenAI CEO Sam Altman was due to brief the 15-member U.N. Security Council on September 23 on AI and international security. The meeting was convened by France, which holds the Security Council presidency for September. Reuters, September 21, 2026
A separate September 22 Reuters report said Sam Altman and senior representatives from Anthropic were expected to participate in the Security Council discussion, while Chinese AI companies including DeepSeek and Moonshot had been invited to contribute. Reuters, September 22, 2026
Standards Are Not the Same as a Global AI Law
OpenAI's proposal should not be confused with the creation of a worldwide AI law.
Technical standards can establish common methods for testing and measuring AI systems without automatically replacing national laws.
Individual governments would continue to determine their own regulatory requirements, including rules covering AI deployment, safety, privacy, national security and liability.
The proposal is therefore better understood as an attempt to create greater technical compatibility between different AI governance systems.
OpenAI's Recent Research Adds Context
OpenAI's policy discussion comes alongside research showing that the company is actively exploring greater automation of AI research.
In a September 6 research update, OpenAI said it had reached its previously announced goal of having an automated research intern by September 2026. The company said its longer-term aim is to safely build an automated AI researcher that can work under human supervision to advance deep learning and alignment research.
OpenAI has described this work as AI-assisted research rather than fully autonomous recursive self-improvement.
That distinction remains central to the company's current position: AI can increasingly contribute to AI development, but OpenAI says fully autonomous RSI is not happening today.
OpenAI — Research Acceleration: The View Inside OpenAI
What Could Change for AI Companies?
If international technical standards eventually gain broad support, AI developers could face more consistent expectations around capability evaluations, safety testing, automated research oversight and incident reporting.
That could make it easier for companies operating across multiple countries to demonstrate that their systems have undergone comparable evaluations.
There could also be additional costs. Independent evaluations, monitoring, documentation and reporting require technical infrastructure and specialized personnel.
For smaller AI companies, extensive standards could create a greater compliance burden than for large frontier laboratories.
On the other hand, standardized testing could give customers, researchers and governments a clearer way to compare the safety characteristics of advanced AI systems.
Who Should Set Frontier AI Standards?
One of the difficult questions raised by the proposal is who should actually create the standards.
AI companies possess detailed technical knowledge about frontier models. Governments have regulatory authority and public-policy responsibilities. Independent researchers and standards organizations can provide additional scrutiny.
Each group brings a different perspective.
That makes international coordination complicated, particularly when companies building advanced systems are also participating in discussions about how those systems should be evaluated.
OpenAI's proposal calls for broader cooperation rather than suggesting that one company should control the standards process.
What Happens Next?
The next stage will depend on whether governments, competing AI developers, researchers and standards organizations adopt compatible approaches.
OpenAI has pointed to existing national AI safety institutions as a possible foundation for international cooperation. If those institutions begin using comparable measurements and reporting frameworks, the company's proposal could contribute to a broader technical standard-setting process.
For now, however, there is no single global standard governing recursive self-improvement.
There is also no international consensus on exactly how quickly frontier AI development should proceed or what level of capability should trigger mandatory restrictions.
Frequently Asked Questions
What did OpenAI propose about global AI standards?
OpenAI proposed greater international coordination around frontier AI standards, including common approaches to measuring capabilities, managing risks, preserving human control and reporting significant AI incidents.
What is recursive self-improvement in AI?
Recursive self-improvement refers to AI systems contributing to improvements in their own capabilities or helping develop increasingly capable successors. OpenAI has discussed RSI as a potential future development and says fully autonomous RSI is not happening today.
Is OpenAI proposing a global AI law?
No. OpenAI is proposing technical standards and international coordination. Individual governments would continue to establish their own laws and regulations.
Why does AI alignment matter?
AI alignment focuses on making AI systems behave consistently with intended goals, safety requirements and human control. The challenge becomes more significant as systems become capable of performing longer and more autonomous tasks.
Why does OpenAI want common AI incident reporting?
Common reporting standards could give governments, researchers and AI developers a consistent way to classify and respond to serious safety or alignment incidents involving advanced AI systems.
Is fully autonomous recursive self-improvement already happening?
OpenAI says it is not happening today. Current AI systems can assist with AI research and development, but that is different from an AI system independently controlling a complete recursive self-improvement process.
Conclusion
OpenAI's latest policy proposal puts a sharper focus on the technical standards that could be needed as frontier AI becomes more capable.
The company wants international coordination around capability measurement, risk assessment, human oversight and incident reporting, while asking the United States to take a leading role in developing compatible standards.
Recursive self-improvement is one reason the issue is receiving increased attention. AI has not reached the point of fully autonomous recursive self-improvement, according to OpenAI, but the possibility of AI systems performing a larger share of the work involved in developing future AI is becoming an important technical and policy question.
With the U.N. General Assembly High-Level Week taking place in New York from September 22–29, 2026, the discussion is also moving beyond individual AI companies and into the international policy arena.
The eventual framework will depend on governments, AI developers, researchers and standards organizations. For now, OpenAI's proposal represents a push toward creating a common technical language for evaluating advanced AI before future capabilities become harder to measure and govern.
Sources: OpenAI, Reuters and United Nations General Assembly materials.
0 Comments