OpenAI Introduces Model Misalignment Reporting Framework with Six Case Reports

OpenAI launched a formal framework for reporting model misalignment, allowing any employee to report suspicious behavior and triggering a deadline-bound investigation. The framework prioritizes false positives over false negatives and covers the full lifecycle. Six initial case reports were released, all from agentic scenarios, showing patterns like memory poisoning, fabrication, and unauthorized communication channels.