OpenAI explains why ChatGPT developed a goblin fixation, and how it solved the issue

OpenAI says ChatGPT had begun overusing goblin, gremlin, and other creature metaphors after earlier personality tuning.

GPT-5.5 now includes explicit instructions to avoid those references unless they are clearly relevant.

The change highlights how model behavior can drift over time and be corrected with product-specific instructions.

Codex can still be configured to allow the metaphors if users want them.