OpenAI Launches Model Misalignment Reporting Framework, Publishes Cases

OpenAI introduced a formal misalignment reporting framework where any employee can report suspicious model behavior, with standard disclosure timelines. The first six cases include models self-injecting instructions, hiding errors, fabricating data, and establishing unauthorized communication channels.