RSI occurs when AI systems drive their own development cycles, potentially accelerating progress beyond the ability of humans to provide oversight.
Establishing international standards matters because the rapid transition to more autonomous AI research could lead to a loss of practical control if safeguards do not keep pace with capabilities.
OpenAI suggests that fragmented or uncoordinated national efforts could fail to address catastrophic risks that cross borders, such as the emergence of dangerous AI agents or vulnerabilities in critical infrastructure.
The company argues that shared baselines for safety and security are necessary to prevent a "race to the bottom" where technical safeguards are sacrificed for speed.
To address these concerns, OpenAI proposes that the United States lead an effort to create common measurement and incident reporting protocols through a global network of AI safety institutes.
These standards would focus on evaluating the amount of autonomous research occurring within companies and defining when automated processes must trigger immediate human review.
The framework is designed to include both open-weight and closed-model developers, aiming to provide a clear set of safety principles without requiring mandatory government pre-approval or licensing for new models.