OpenAI Calls for Global Standards on AI Alignment and Recursive Self-Improvement

OpenAI published a set of proposals on Monday outlining safety and security frameworks for frontier artificial intelligence, placing heavy emphasis on alignment research and recursive self-improvement, as reported by CNBC Business (https://www.cnbc.com/2026/09/21/open-ai-alignment-rsi.html).

In a company blog post, the ChatGPT creator argued that technical safety research must match the pace of capability gains. OpenAI stated that navigating this transition safely requires alignment research to keep pace with capabilities so that systems remain aligned with human values and under human control.

To establish enforceable safeguards, OpenAI recommended international cooperation to craft frontier AI standards, proposing that global policymakers build on the work of existing AI safety institutes. The company said these technical standards should specifically target frontier models and developers, alongside benefit-risk management frameworks for automated AI researchers capable of recursive self-improvement.

Recursive self-improvement, or RSI, involves foundation models that can autonomously upgrade and iterate on their own code without human intervention. While developers view RSI as a path to rapid capability gains, technologists have warned that creators could lose practical control of systems or trigger unintended consequences as models become increasingly complex and ubiquitous across the internet.

OpenAI noted that fully autonomous RSI is not occurring today and stated that industry actors should not pursue it unless and until it can be managed safely. The company warned that uncontrolled RSI could leave humans unable to oversee research processes they no longer understand.

The proposals follow a similar framework released last week by competitor Anthropic. Industry scrutiny over frontier safety practices has sharpened since researcher Jacob Coxon, who previously worked at both OpenAI and Anthropic, resigned nearly two weeks ago and claimed the labs were gambling with human lives.