OpenAI proposes international standards for frontier models

OpenAI proposed an international effort to develop technical standards for frontier AI, including systems that help improve subsequent AI systems. The company called for the United States to lead coordination through public institutions, national industry bodies and the existing network of AI safety institutes.

The proposal identifies inconsistent evaluations, fragmented incident reporting and uneven technical expertise as obstacles to managing risks across borders. It suggests common measures of research autonomy, thresholds for human review, and shared classifications for alignment or automated-research incidents.

OpenAI says the technical standards should cover both open and closed models and should not themselves create licensing or mandatory approval requirements. Governments would decide how to incorporate them into domestic law. The announcement is a policy proposal, with the participating institutions, detailed standards and implementation still to be agreed.