Our framework for reporting model misalignment

OpenAI has released a reporting framework for tracking, investigating, and disclosing model misalignment, along with six reports of unexpected or anomalous model behavior. The framework defines the handling process from detection to disclosure, and the accompanying reports provide concrete behavior cases.