A peculiar directive has been uncovered within the system prompt for OpenAI’s Codex Command Line Interface (CLI), specifically targeting the latest GPT-5.5 model. It explicitly instructs the model to "never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query."
This explicit operational warning was made public last week as part of the most recent open-source code for Codex CLI, which OpenAI posted on GitHub. The prohibition appears twice within a set of over 3,500 "base instructions" for the recently released GPT-5.5. These instructions also include more standard reminders, such as not to "use emojis or em dashes unless explicitly instructed" and to "never use destructive commands like ‘git reset --hard’ or ‘git checkout --’ unless the user has clearly asked for that operation."
Notably, separate system prompt instructions for earlier models, contained within the same JSON file, do not feature this specific prohibition against mentioning goblins and other creatures. This omission suggests that OpenAI is addressing a novel issue that has emerged with its latest model release. Anecdotal evidence circulating on social media corroborates this, with some users reporting GPT’s tendency to focus on goblins in conversations entirely unrelated to the topic at hand.
OpenAI employee Nick Pash, who works on Codex, took to social media to state that this "isn’t a marketing gimmick" designed to generate buzz around GPT-5.5 and Codex. Despite this, some OpenAI executives have embraced the humor surrounding the system prompt as the news spread. OpenAI CEO Sam Altman, for instance, humorously remarked on social media, "Feels like codex is having a ChatGPT moment. I meant a goblin moment, sorry."