OpenAI Codex system prompt includes a 'no goblin mentions' instruction

The latest system prompt for OpenAI Codex includes an instruction not to mention creatures such as goblins, gremlins, raccoons, trolls, ogres, and pigeons in irrelevant contexts. This phrase appears twice in the over 3,500-word base instructions for GPT-5.5, and was confirmed in the Codex CLI code published on GitHub.

• The prompt for previous models did not contain this prohibition, suggesting it's a new response added for recent models.
• Some users complained on social media that GPT keeps bringing up goblins regardless of the conversation.
• Nick Pash said this wasn't a marketing gimmick, and Sam Altman made a related joke about it.
• Users have started creating plugins, forks, and AI skills to work around this, and Pash also mentioned the possibility of a future goblin mode toggle.

The article also compares this situation to xAI Grok's past obsession with white genocide. At the time, xAI explained that an unauthorized system prompt modification was the cause, and it subsequently began publishing its system prompts on GitHub.

At the same time, the prompt is designed to make the model feel intelligent, playful, curious, and deep. The direction aims to give users the sense of encountering another entity, rather than just a simple tool.