OpenAI Calls for Global Standards for AI That Can Improve Itself

OpenAI has called for the United States to lead an international effort to establish technical standards for the most advanced AI systems, including systems capable of recursively improving themselves.

The proposal comes as world leaders gather in New York for the UN General Assembly and as governments face growing questions over how increasingly capable AI should be governed.

OpenAI wants countries to develop common standards for measuring frontier AI capabilities, maintaining human oversight and responding to safety incidents.

A central focus is recursive self-improvement, or RSI, in which AI increasingly contributes to the research and development needed to produce more capable generations of AI.

OpenAI says fully autonomous recursive self-improvement is not happening today and should not be pursued unless it can be done safely. However, AI systems are already assisting with parts of the research used to develop subsequent models.

As more of that work becomes automated, OpenAI says the pace of AI development could accelerate substantially.

The company is proposing common methods for measuring how much autonomous AI research is taking place, standards for human oversight and shared protocols for classifying and reporting incidents.

OpenAI argues that incompatible national rules could make it harder for governments and AI laboratories to compare evidence and respond collectively to emerging risks.

It wants the US to take a leading role in developing the standards alongside other countries, while maintaining national authority over how they are ultimately incorporated into law.

The proposal also calls for secure international channels through which governments could exchange information about vulnerabilities, emerging threats and national-security concerns. OpenAI has specifically supported dialogue between the United States and China on AI safety.

The initiative comes amid disagreement over whether the development of increasingly powerful AI should be slowed.

President Donald Trump has opposed a general slowdown, with his administration concerned that restricting US development could give China a technological advantage.

OpenAI’s position stops short of calling for a predetermined reduction in development speed. Instead, it argues that safety and alignment work must remain ahead of advancing capabilities and that development should slow when necessary.

The issue will move directly onto the international stage this week.

OpenAI CEO Sam Altman is scheduled to brief the UN Security Council on Wednesday during an open meeting on artificial intelligence and international security. His remarks are expected to address international coordination and shared AI safety standards.

The debate is therefore moving beyond broad calls for “AI safety” towards a more specific question: what international safeguards should apply if AI begins playing a substantial role in building increasingly capable versions of itself?

Sources

• Reuters: OpenAI calls for US to take lead on global AI technical standards
• OpenAI: The AI policy window