A shared playbook for trustworthy third party evaluations

In three linesOpenAI publishes guidance for third-party AI evaluations, covering assessment of model capabilities, safeguards, and validity for frontier systems.

Summary generated by Claude — human-verified