OpenAI’s system prompt for the Codex CLI for the latest GPT model contains a confusing and repeated warning to “never mention goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is completely and unequivocally relevant to the user’s query.”
There was an open transaction alert made public last week As part of the latest open source code for Codex CLI, which is OpenAI Hosted on GitHub. The ban is repeated twice in the recently released GPT-5.5 set of more than 3,500 words of “core guidelines”, along with more anodyne reminders to “not use emojis or dashes unless explicitly instructed” and “never use destructive commands like ‘git reset –hardout'” or check that the user has clearly verified the operation.
For previous models in the same JSON file, there is no specific prohibition against specifying goblins and other creatures in the separate system specification instructions, which indicates that OpenAI is tackling a new problem that appeared in the latest model release. An anecdote proof about social media shows some users complain About GPT’s recent tendency to focus on goblins in completely unrelated conversations.
Nick Pash of OpenAI working on Codex, insists on social media This is “not a marketing ploy” to get people talking about GPT-5.5 and Codex. But that hasn’t stopped some OpenAI executives from indulging in jokes about how quickly the system has spread. “It’s like codex is having a ChatGPT moment. I meant bold moment, sorry,” OpenAI CEO Sam Altman he wrote on the social network on Wednesday morning.
