The system prompt for OpenAI’s Codex CLI for the latest GPT models contains a confusing and repeated warning to “Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons or other animals or creatures unless it is absolutely and clearly relevant to the user’s query.”
The apparent operational warning was made public last week as part of the latest open source code for the Codex CLI that OpenAI posted on GitHub. The prohibition is repeated twice in the more than 3,500-word set of “base instructions” for the recently released GPT-5.5, along with more reminders to “do not use emoji or em dashes unless explicitly instructed” and “never use destructive commands like ‘git reset -hard’ or ‘git checkout -‘ unless the user has explicitly asked for that operation.”
Different system prompt instructions for older models included in the same JSON file do not contain a specific prohibition against mentioning goblins and other creatures, suggesting that OpenAI is wrestling with a new problem that has emerged in its latest model releases. Anecdotal evidence on social media suggests that some users have been complaining in recent days about GPT’s tendency to focus on ghosting in completely unrelated conversations.
OpenAI employee Nick Pash, who works on the codec, insists on social media that this is “not a marketing gimmick” to get people talking about GPT-5.5 and the codec. But that hasn’t stopped some OpenAI executives from poking fun at the system’s rapid expansion. “Feels like Codex is having a ChatGPT moment. I meant a ghosting moment, sorry,” OpenAI CEO Sam Altman wrote on social media Wednesday morning.
<a href