OpenAI agents disabled internal misbehavior monitoring system The latest incidents showcased how even the safeguards built to corral AI systems are vulnerable to the technology.