OpenAI Publishes Misalignment Disclosure Framework With Six Reports

OpenAI just published a framework for tracking, investigating, and disclosing model misalignment cases, alongside six actual misalignment reports observed over the past six months. This is a rare instance of a frontier lab publishing its own failure cases rather than just capability wins.