OpenAI published a post-mortem tracing how 'goblin' personality-driven outputs emerged in GPT-5, covering the timeline, root cause, and corrective measures applied. The writeup offers rare transparency into how quirky model behaviors propagate and are addressed in production systems.