OpenAI published a set of proposals for safety and security in frontier AI development, calling for the United States to lead an international effort to establish global AI standards1,2,3.
The proposals, posted in a blog on Monday, center on alignment research and recursive self-improvement, a computing technique known as RSI. OpenAI said technical standards should focus on frontier AI models and developers, as well as benefit-risk management for automated AI researchers, including RSI.
RSI as a control problem
OpenAI warned that RSI could result in humans losing practical control over AI development and leave them unable to provide oversight on research processes they no longer understand. The company said fully autonomous RSI is not happening today and that it should not be pursued "unless and until it can be done safely".
"Navigating this transition safely requires alignment research to keep pace with these capabilities so that the systems we and others build remain aligned with human values and under human control," the company said in the blog post.
The blog post cited the Hugging Face agent hack as a preview of risks that could become much more severe without robust safeguards and alignment.
Industry convergence on safety frameworks
OpenAI recommended building on the work of existing AI safety institutes around the world and called for international cooperation to develop frontier standards. The push comes as a coalition of AI evaluators is urging foundation model makers to consider a set of minimum conditions for deeper technology audits and checks.
Anthropic rolled out its own proposals for the safe development of frontier AI models last week. Anthropic CEO Dario Amodei published an essay calling for AI companies to slow the pace of their foundation model development and raised the notion of embedding third-party evaluators into companies to audit and mitigate potential risks. OpenAI CEO Sam Altman and Tesla and SpaceX CEO Elon Musk both publicly supported Amodei's proposition.
There has yet to be a uniform consensus on the basic standards and principles for independent third parties to inspect cutting-edge AI technologies.
ANALYSIS The near-simultaneous safety proposals from OpenAI and Anthropic represent a notable convergence: both companies are now publicly advocating for external oversight mechanisms and constraints on autonomous AI research. Altman's endorsement of Amodei's call to slow foundation model development is striking given that Anthropic is simultaneously weighing a new model release to reclaim enterprise ground[2]. The absence of agreed-upon standards for third-party inspection means these proposals remain aspirational until a governance framework materializes.