
A document posted by OpenAI on Github as part of the open-sourcing for Codex CLI, OpenAI’s most recent flagship coding agent, contains what looks like the entire system prompt for GPT-5.5 in a coding context. And it seems to correct for the model’s past addiction to talking about whimsical creatures both natural and supernatural.
Here’s the relevant section (emphasis added):
“provide the highest-signal context instead of describing everything exhaustively.\n- Tone of your final answer must match your personality.\n- Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query.”
Evidently this point is so important, the developers bring it up again a bit later:
“For example, never use platitudes like \”I will do <this good thing> rather than <this obviously bad thing>\’, \’I will do <X>, not <Y>\’.\n- Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query.”
It’s not clear why this matters so much. If these were, for instance, canary words inserted into the system prompt as a way to monitor for prompt injection attacks, they would probably be more random, rather than seemingly a whole category of animals.
OpenAI did not reply to a request for comment on Tuesday night.
A Google employee named Barron Roth posted what appears to be a search of his chat logs with some of his GPT-5.5-powered Openclaw agents, showing that at least one had a history of inserting the word “goblin” into messages to the user multiple times in a single day. To my admittedly untrained eye, GPT-5.5 seems to be using it in place of a word like “thingy.”
this explains gpt5.5’s goblin adoration in openclaw and why @pashmerepat is making us move to the codex harness 🤣🤣🤣 pic.twitter.com/N20UhQDOK5
— Barron Roth (@iamBarronRoth) April 28, 2026
Nick Pash, who works on Codex at OpenAI, partially confirmed that Roth had hit on the nature of the problem, writing to him on X “this is indeed one of the reasons.”
It looks like other X users also noticed that Codex had taken the goblin talk a little too far.
On Tuesday, this goblin issue became a meme, with users suggesting that some sort of “Goblin Mode,” could be toggled on and off.
Pash eventually added his own post about Goblin Mode:
Alright fine. pic.twitter.com/6RV1SHx78C
— pash (@pashmerepat) April 28, 2026
If this is giving you flashbacks to a year ago when OpenAI was all about Studio Ghibli memes, you’re not alone. Some dared to assert that OpenAI might be doing all this goblin stuff for attention at what is, after all, a somewhat troubled moment for the company.
But according to another post by Pash, “it really isn’t a marketing gimmick.”