OpenAI explains why ChatGPT suddenly loved goblins

OpenAI said ChatGPT’s goblin and gremlin references were traced to reinforcement effects tied to its now-removed “nerdy” personality mode.

The company retired that mode and updated newer models so they stop making those mentions unless they are actually relevant.

The fix follows a GPT-5.1-era bug where a small stylistic behavior spilled into broader responses.

The incident underscores how hard it can be to steer AI personalities and prevent odd, unwanted output.