OpenAI moves third-party safety assessments to earlier stages of model development
OpenAI will expand third-party participation in safety testing during the early stages of model training, evaluation, and deployment, and is in discussions with organizations including METR and Redwood Research.