The system timely for OpenAI’s Codex CLI consists of a difficult and repetitive caution for the most current GPT design to “never ever discuss goblins, gremlins, raccoons, giants, trolls, pigeons, or other animals or animals unless it is definitely and unambiguously pertinent to the user’s inquiry.”

The specific functional caution was revealed recently as part of the most recent open source code for Codex CLI that OpenAI published on GitHub. The restriction is duplicated two times in a 3,500-plus word set of “base directions” for the just recently launched GPT-5.5, along with more anodyne pointers not to “utilize emojis or em dashes unless clearly advised” and to “never ever utilize damaging commands like ‘git reset– difficult’ or ‘git checkout–‘ unless the user has actually plainly requested that operation.”

Different system timely guidelines for earlier designs included in the very same JSON file do not include the particular restriction versus pointing out goblins and other animals, recommending OpenAI is combating a brand-new issue that has actually appeared in its most current design release. Anecdotal proof on social networks reveals some users grumbling about GPT’s fondness for concentrating on goblins in totally unassociated discussions in current days.

OpenAI worker Nick Pash, who deals with Codex, demands social networks that this “isn’t a marketing trick” to get individuals discussing GPT-5.5 and Codex. That hasn’t stopped some OpenAI executives from leaning into the joke as word of the system timely spread. “Feels like codex is having a ChatGPT minute. I indicated a goblin minute, sorry,” OpenAI CEO Sam Altman composed on social networks Wednesday early morning.