OpenAI calls for global standards on frontier AI and recursive self‑improvement

The essay proposes that the United States lead an international effort to create technical standards for frontier AI, using the Center for AI Standards and Innovation (CAISI) and existing networks such as the International Network for Advanced AI Measurement, Evaluation, and Science (launched in 2024). OpenAI suggests standards covering RSI evaluation, human‑oversight triggers, and incident classification, and it points to the recent Hugging Face incident as a preview of risks without robust safeguards. Collaboration with bodies like ISO, the Frontier Model Forum, the Agentic AI Foundation, the Open Secure AI Alliance, and the Appia Foundation would help keep standards transparent and avoid favoring any single company or country.

OpenAI believes that coordinated standards can keep alignment research ahead of capability growth, protect critical infrastructure, and enable a competitive yet safe AI ecosystem worldwide. The piece calls for democratic input, public‑private partnerships, and dialogue—particularly with China—to shape a balanced global framework.