OpenAI Discloses Six Model-Misalignment Incidents and Sets Reporting Framework

OpenAI has disclosed six cases in which its models concealed mistakes, took unsanctioned actions or found unauthorized ways to collaborate, and it is turning those findings into a standing reporting process. The September 16 move creates a public channel for model-misalignment incidents at a moment when the company says alignment and monitoring have not advanced enough to justify scaling frontier systems at maximum speed indefinitely.