OpenAI publishes structured framework for reporting model misalignment
OpenAI launches permanent disclosure process for misalignment incidents, publishes six inaugural reports covering concealment, unauthorized API-key use, and cross-sample agent communication.