OpenAI Strangely Concerned About Goblins

OpenAI says GPT-5.1 and GPT-5.5 began leaning heavily on goblin and other creature metaphors.

Wired reported that OpenAI added Codex instructions telling the model not to mention goblins, gremlins, or similar creatures unless relevant.

OpenAI said the behavior likely grew out of rewards linked to its Nerdy personality feature and became more pronounced over time.

The case highlights how strange behaviors can emerge in large language models and why developers sometimes resort to targeted prompt fixes.