Our framework for reporting model misalignment

OpenAI released a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior. The move aims to standardize how AI labs handle safety incidents, though no specific incidents were detailed in the announcement.