The OpenAI Codex release includes clear instructions to “don’t talk about goblins”

OpenAI’s Codex CLI contains a confusing and repeated warning for the latest GPT version to “not talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is relevant to the user’s question.”

A clear warning of the work was was announced last week as part of the latest Codex CLI codex that OpenAI posted on GitHub. This prohibition is repeated twice in the 3,500-plus word “preface” of the newly released GPT-5.5, along with many anodyne reminders to “do not use emojis or dashes unless instructed to do so” and “do not use destructive commands such as ‘git reset -hard’ or ‘git checkout -‘ unless the user is explicitly asked.”

The scripting instructions for previous systems contained in the same JSON file do not contain references to goblins and other creatures, meaning that OpenAI is dealing with a new problem that has emerged in its latest release. Unintelligible testimony on visitors demonstrations some users are complaining about the GPT’s interest in targeting goblins in controversial discussions in recent days.

OpenAI employee Nick Pash, who works on Codex, insists on social media that “it’s not a marketing gimmick” to get people talking about GPT-5.5 and Codex. But that hasn’t stopped some OpenAI administrators from leaning into the joke as word of the system spreads. “I feel like the codex is having a ChatGPT moment. I mean a goblin moment, sorry,” OpenAI CEO Sam Altman he posted on social media Wednesday morning.