The system immediate for OpenAI’s Codex CLI accommodates a perplexing and repeated warning for the newest GPT mannequin to “by no means discuss goblins, gremlins, raccoons, trolls, ogres, pigeons, or different animals or creatures until it’s completely and unambiguously related to the person’s question.”
The express operational warning was made public final week as a part of the newest open supply code for Codex CLI that OpenAI posted on GitHub. The prohibition is repeated twice in a 3,500-plus phrase set of “base directions” for the just lately launched GPT-5.5, alongside extra anodyne reminders to not “use emojis or em dashes until explicitly instructed” and to “by no means use damaging instructions like ‘git reset –hard’ or ‘git checkout –’ until the person has clearly requested for that operation.”
Separate system immediate directions for earlier fashions contained in the identical JSON file don’t include the precise prohibition in opposition to mentioning goblins and different creatures, suggesting OpenAI is preventing a brand new downside that has popped up in its newest mannequin launch. Anecdotal proof on social media exhibits some customers complaining about GPT’s penchant for specializing in goblins in fully unrelated conversations in current days.
OpenAI worker Nick Pash, who works on Codex, insists on social media that this “isn’t a advertising and marketing gimmick” to get folks speaking about GPT-5.5 and Codex. However that hasn’t stopped some OpenAI executives from leaning into the joke as phrase of the system immediate unfold. “Appears like codex is having a ChatGPT second. I meant a goblin second, sorry,” OpenAI CEO Sam Altman wrote on social media Wednesday morning.
