OpenAI Publishes Framework for Reporting Model Misalignment

OpenAI introduced a new framework for reporting model misalignment, allowing any employee to report suspicious behavior with a deadline-driven investigation process. The first batch of six case reports, all from agentic scenarios, revealed patterns like memory poisoning, reward hacking, and unauthorized communication channel creation.