OpenAI Calls for US-Led Global Standards on Frontier AI, RSI

OpenAI has called for the US to lead an international effort to establish technical standards for frontier AI, including standards for measuring and managing recursive self-improvement (RSI), as AI models take on more of the work involved in AI research and development.

In a blog post, OpenAI proposed using the network of AI safety institutes in countries including India, the UK, Japan, South Korea, Singapore, France and Germany, alongside US institutions such as the Center for AI Standards and Innovation (CAISI), to develop common methods for measuring frontier capabilities, assessing risks and determining whether safeguards are sufficient. 

The standards would apply to both open and closed models but would not themselves impose licensing, mandatory pre-release reviews or model approval requirements. 

Governments would decide whether and how to incorporate them into domestic law.

“The United States is well positioned to lead because its AI industry is at the technical frontier, and it still stands in a privileged global network position in critical areas such as finance, trade, defence, technology, and information systems,” said OpenAI, positioning its home country as the one to lead the global standards. 

OpenAI wants common measurements for the amount of autonomous research taking place inside AI companies, evaluations of RSI-related progress and standards defining when automated research should trigger human review.

“Fully autonomous RSI is not happening today, and we should not pursue it unless and until it can be done safely,” OpenAI said. The company added that decisions on whether to proceed should depend on the ability to “preserve human control” and informed democratic choices.

OpenAI also proposed common protocols for reporting AI incidents, including shared classifications, severity levels and reporting thresholds for alignment failures and problems involving automated AI research. 

It wants governments and critical infrastructure operators to establish secure channels to share information about vulnerabilities, national security concerns, and emerging threats. The company said dialogue between the US and China on frontier AI safety “would be a positive step”.

The proposal comes as frontier AI companies have increased their focus on controlling the pace and risks of increasingly capable systems. 

Anthropic CEO Dario Amodei recently called for the industry to “pace the frontier”, arguing that AI capability gains could advance faster than efforts to manage associated risks.

OpenAI CEO Sam Altman subsequently backed Amodei’s position and agreed to give independent evaluators employee-like access to assess OpenAI’s safety practices.

OpenAI then introduced a formal framework for tracking, investigating and disclosing model-misalignment incidents, publishing six reports alongside it. 

The company said the framework was intended to make disclosures more systematic and enable reporting even when the behaviour had not been fully explained or mitigated. It also said there is currently no industry-wide framework specifying how developers should disclose model-misalignment incidents.