If AI achieves recursive evolution, can humans still control it? OpenAI CEO Sam Altman recently posted that people outside AI labs should have genuine input and clear methods to assess whether AI is developing safely. OpenAI subsequently released a proposal advocating for the United States to lead the creation of international AI technical standards. What exactly is this proposal aiming to regulate? 1. When AI begins developing the next generation of AI, humans must keep pace. The article focuses on “recursive self-improvement” (RSI): AI participates in developing more powerful AI, which in turn continues advancing development—a process that could accelerate rapidly. OpenAI fears that as automation increases, humans may gradually lose the ability to understand the research process and thus lose effective oversight. The company explicitly states that fully autonomous RSI has not yet occurred and should not be pursued until safety can be guaranteed. 2. Establishing comparable standards to assess safety. How much research is being conducted autonomously by AI? Which scenarios must be immediately reviewed by humans? In the event of an incident, how should it be classified, reported, and addressed? OpenAI hopes that countries and companies adopt common standards so that safety evidence can be compared across organizations, allowing one company’s failures to help others avoid similar mistakes. 3. Setting rules while preserving space for new entrants and open models. Sam emphasized that standards should help prevent concentration of power. The proposal also calls for input from open-model developers, independent experts, and academia to ensure rules do not favor specific companies or make it harder for newcomers to enter the market. These standards are not intended as model licenses or pre-release approval requirements; individual countries may decide how to incorporate them into their own laws. This remains a proposal for now. The extent of influence for those outside labs and the specifics of how common standards will be developed still require further elaboration.