On September 21, 2026, OpenAI released a set of proposals aimed at establishing international standards for the safe development of frontier artificial intelligence. The company stressed the need for alignment research to ensure that AI systems align with human values and remain controllable.
OpenAI's blog post highlighted the significance of recursive self-improvement (RSI), which allows AI models to enhance themselves autonomously, raising concerns about potential loss of control over these technologies.
The company called for collaboration with existing AI safety institutes globally and recommended that technical standards focus on frontier AI models and risk management for automated AI researchers. OpenAI cautioned that pursuing fully autonomous RSI without adequate safety measures could lead to humans losing oversight of AI development.
This announcement comes in the wake of similar proposals from rival company Anthropic, reflecting a broader industry response to warnings about AI's potential risks. The debate intensified following Jacob Coxon's resignation from Anthropic and OpenAI, where he expressed concerns about the industry's approach to AI safety