OpenAI is calling for international technical standards for advanced AI. These standards should primarily clarify systems that conduct increasingly autonomous AI research and can contribute to future generations of models. The United States should take the lead in this effort.
The focus is on recursive self-improvement (RSI). This involves using AI systems to develop new and better AI systems. According to OpenAI, as humans take fewer steps in that research process, development can accelerate significantly.
According to OpenAI, fully autonomous RSI does not yet exist and should not be pursued until it is clear that it can be done safely. Without sufficient oversight, humans could lose practical control over AI development, for example, because research processes become too complex to understand or evaluate.
Different rules by country
OpenAI views international standards as a way to mitigate such risks. Different definitions, evaluation methods, and reporting requirements by country make it more difficult to compare AI systems and jointly address cross-border risks. The standards should apply to both open and closed models.
The company does not advocate for an international licensing system. The agreements should form a common basis for measuring capabilities, assessing risks, and evaluating safety measures. National governments will determine for themselves whether and how to incorporate the standards into legislation.
This effort can build on the international network of AI safety institutes. Organizations and partnerships such as ISO, the Frontier Model Forum, and the Open Secure AI Alliance can also play a role.
Establishing human oversight
The agreements must, among other things, specify how progress toward self-improving AI is measured and when automated AI research must be immediately reviewed by humans.
In addition, OpenAI wants common rules for classifying, recording, and reporting incidents. Critical infrastructure operators and governments should establish secure communication channels to exchange information about vulnerabilities, threats, and security measures. OpenAI explicitly states that cooperation between the United States and China is desirable in this regard.
The proposal comes at a time when AI safety is high on the international political agenda. Altman will address the UN Security Council on Wednesday regarding AI and international security. At the same time, the Trump administration is rejecting calls to slow down the development of AI, in part out of fear that China would benefit from it, Reuters reports.
The U.S. must take the lead
OpenAI sees a leading role for the United States because of its strong AI industry and the country’s position in international networks for technology, trade, finance, and defense, among other areas.
According to OpenAI, international competition is therefore not just about who develops the most powerful models. Establishing widely accepted technical standards and organizing international cooperation can also determine how AI is applied worldwide.