Our framework for reporting model misalignment

OpenAI’s framework for tracking, investigating, and disclosing model misalignment, with six reports on unexpected or concerning behavior from the last six months.