ChatGPT's goblin obsession may be hilarious, but it points to a deeper problem in AI training

OpenAI found a training bug in ChatGPT’s “Nerdy” personality that overused goblin-like metaphors.

The faulty reward signal spread the behavior beyond that mode and affected models including GPT-5.1, GPT-5.5, and Codex.

OpenAI disabled the personality, removed the signal, filtered creature terms, and added instructions to avoid the metaphors.

The episode highlights how small training incentives can create broad, unintended quirks that affect reliability and product quality.