OpenAI Proposes US-Led Global Technical Standards for Frontier AI

OpenAI on September 21, 2026 published “Building standards for the next phase of AI,” a Global Affairs post calling for the United States to lead an effort with countries around the world to develop global technical standards for frontier AI, including standards for recursive self-improvement.

The proposal sits within three mission goals OpenAI said were outlined by Sam Altman and Jakub Pachocki: building an automated AI researcher, iterating with it on the alignment problem, and finding ways for people to remain part of the self-improvement loop; delivering the scientific and economic benefits that very intelligent machines enable; and empowering people individually with a personal AGI. Keeping AI progress safe and beneficial, OpenAI said, depends on both advances in alignment research and shared standards guiding development across labs and countries.

OpenAI’s Position on Recursive Self-Improvement

The degree of human supervision in automated AI research can vary, the post explains. When AI systems handle more of the work of producing successive generations of AI, they can drive a process of recursive self-improvement, or RSI, even with people still involved, and OpenAI said progress could accelerate rapidly as the process grows more automated.

OpenAI stated its belief that automated AI research will produce models that directly improve people’s lives and lower the cost of advanced intelligence, and that an automated AI researcher can also serve as an automated AI safety researcher. At the same time, the company stated that fully autonomous RSI is not happening and should not be pursued unless and until it can be done safely. Any decision on whether and how to proceed, OpenAI said, must depend on the ability to preserve human control and on informed democratic choices about benefits and risks.

Handled without appropriate care and caution, RSI could leave humans without practical control over AI development and unable to oversee research processes they no longer understand, the company said. OpenAI described the Hugging Face Incident it previously disclosed, which it said was not a direct result of RSI, as an early look at the kinds of risks that could grow far more severe without robust safeguards and alignment.

Three Challenges Behind the Standards Push

International standards for safety and security practices in frontier AI development could matter as much for pacing the frontier as alignment research itself, OpenAI argued, because they give governments and labs shared definitions of what counts as high-quality evidence and common baselines for how rigorous technical safeguards must be. The post identifies three challenges such standards are meant to solve.

The first is fragmentation: evaluations, reporting requirements, and incident definitions that conflict between countries would make cross-border evidence harder to compare, emerging capabilities harder to gauge, and shared risks harder to answer. The second is collective action: countries acting alone can end up with results none of them wants, and OpenAI said RSI could speed up AI research faster than nations can collectively follow progress, gauge risks, and keep meaningful human oversight.

The third is uneven capacity: frontier development activity and expertise are unevenly spread around the world, which the post says worsens the other two problems. These challenges apply to both open and closed models, OpenAI said, adding that any AI lab pursuing automated AI research or other advanced capabilities must take accountability for doing so safely. The company described its rationale for standards as rooted in avoiding concentrated power and producing better practical outcomes, giving stakeholders outside the labs a say in how the technology unfolds.

A Mechanism Built on CAISI and the Safety Institute Network

OpenAI said it expects the concept to evolve substantially over time but identified two aspects as essential. The first is a mechanism that facilitates complementary national and international frontier standards. One route, the post proposes, is to build on the emerging network of AI safety institutes, such as those already established in Australia, Canada, Germany, France, Kenya, Japan, Korea, Singapore, India, and the United Kingdom, and to set standards through the Center for AI Standards and Innovation (CAISI) and national industry bodies. The work would focus on frontier AI models and developers as measured by capability benchmarks, and on benefit-risk management for automated AI research, including RSI.

OpenAI said the United States can build on CAISI’s 2024 creation of the International Network for Advanced AI Measurement, Evaluation, and Science. According to the National Institute of Standards and Technology, CAISI founded that network in November 2024, and its membership spans government bodies from ten countries: Australia, Canada, the European Union, France, Japan, Kenya, the Republic of Korea, Singapore, the United Kingdom, and the United States.

NIST reported that Secretary of Commerce Howard Lutnick gave CAISI a charge in June 2025 to develop guidelines and best practices for measuring and improving the security of AI systems and to help industry develop voluntary standards. CAISI convened network members with industry and technical-organization representatives in San Diego during NeurIPS in December 2025 for exchanges on best practices and open questions in automated AI evaluations, and the network published consensus areas on practices for automated AI evaluations on February 13, 2026.

Standards from this effort would give a shared technical basis for capability measurement, risk assessment, and evaluating whether safeguards are sufficient, OpenAI said. The company said they would not function as licenses, nor as mandatory prerelease review or approval requirements for AI models, and that national governments would decide whether and how to fold them into their own legal systems.

The work should consult open and closed model developers, independent technical experts, and academia, OpenAI said, and the standards should be built transparently and designed not to favor particular companies, countries, or business models, including by making it harder for new entrants or open-weight developers to compete. The post says the approach can borrow lessons from aviation and financial stability, where countries have built common technical standards and trusted cooperation channels without surrendering national authority, and that the effort should work with ISO, the Frontier Model Forum, the Agentic AI Foundation, and the Open Secure AI Alliance, as well as the Appia Foundation, which is developing practical specifications connecting international standards to real-world AI assessments.

Common Measurements, Incident Protocols, and US–China Dialogue

The second essential aspect, OpenAI said, is common measurements and incident reporting protocols for better collective action. Standards could cover how RSI-relevant AI progress and the amount of autonomous research inside an AI company are evaluated, and OpenAI cited its report on research acceleration as an initial contribution. They could also cover human oversight of automated AI research, including which automated research processes should trigger immediate human review, and how alignment and automated-research incidents are classified, tracked, reported, and answered, with shared severity levels and reporting thresholds. OpenAI described its misalignment reporting framework as an early contribution to this work.

Beyond standards, the company said critical infrastructure operators and governments worldwide will need secure communication channels to share national security concerns, emerging vulnerabilities and threats, and best practices in frontier AI safety. OpenAI said dialogue between the United States and China in these areas would be a positive step, and that planned talks come at an opportune moment.

In a closing section, OpenAI argued that pacing AI development is not about holding to a predetermined speed but about keeping alignment research and its deployment ahead of capabilities. In the company’s view, the United States should lead because its AI industry operates at the technical frontier, and strong national governance linked by practical international cooperation offers a route to stronger safeguards, continued innovation, and broad access to the benefits of AI.