OpenAI identified and is addressing a bug in ChatGPT models that caused the AI to spontaneously discuss goblins in conversations where they were not relevant or mentioned by users. The company characterized the issue as one that "crept in subtly," distinguishing it from more obvious model glitches.
The firm did not disclose when the problem emerged, how widespread it was, or which specific ChatGPT versions were affected. OpenAI also did not explain the technical cause behind the aberrant behavior, though such unexplained outputs in large language models typically stem from patterns learned during training data that become activated in certain contexts.
This type of drift, where models generate off-topic or contextually inappropriate content, represents a known challenge in AI development. While the goblin references appear harmless, the incident underscores broader concerns about model reliability and the difficulty of predicting how large language models will behave across diverse user interactions. OpenAI's response suggests the issue has been resolved or is being actively corrected in deployed versions.