TO document published by OpenAI on Github As part of the open source code for Codex CLI, OpenAI’s latest flagship encryption agent, it contains what appears to be the full system message for GPT-5.5 in an encryption context. And it seems to correct the model’s past addiction to talking about capricious creatures both natural and supernatural.
Here is the relevant section (emphasis added):
“provide the highest cue context rather than describing everything exhaustively.\n- The tone of your final answer should match your personality.\n- Never talk about goblins, elves, raccoons, trolls, ogres, doves or other animals or creatures unless it is absolutely and unequivocally relevant to the user’s query..”
Obviously this point is so important that the developers mention it again a little later:
“For example, never use platitudes like “I will.”
instead of \’, \’I will No \’.\north- Never talk about goblins, elves, raccoons, trolls, ogres, doves or other animals or creatures unless it is absolutely and unequivocally relevant to the user’s query.”
It is not clear why this is so important. If these were, for example, canary words inserted into the system prompt as a way to monitor immediate injection attacks, they would likely be more random, rather than seemingly an entire category of animals.
OpenAI did not respond to a request for comment Tuesday evening.
A Google employee named Barron Roth posted what appears to be a search of his chat logs with some of his GPT-5.5 Openclaw agents, showing that at least one had a history of inserting the word “goblin” in messages to the user multiple times in a single day. To my admittedly untrained eye, GPT-5.5 appears to use it in place of a word like “thingy.”
this explains the gpt5.5 goblin worship in openclaw and why @pashmerepat he is making us pass into the codex harness 🤣🤣🤣 pic.twitter.com/N20UhQDOK5
– Barrón Roth (@iamBarronRoth) April 28, 2026
Nick Pash, who works on Codex at OpenAI, partially confirmed that Roth had nailed down the nature of the problem. writing to him in X “this is indeed one of the reasons.”
It seems that other X users also noticed that Codex I had taken the goblin talk too far..
On Tuesday, this elf problem became a meme and users suggestion that some type of “Goblin Mode” could be activated and deactivated.
Pash finally added his own post about Goblin mode:
Very good. pic.twitter.com/6RV1SHx78C
-pash (@pashmerepat) April 28, 2026
If this reminds you of a year ago when OpenAI had to do with Studio Ghibli memesyou are not alone. Some dared to affirm that OpenAI might be doing all this pixie stuff to draw attention to what is, after all, a somewhat turbulent time for the company.
But according to Another post from Pash.“It’s really not a marketing gimmick.”
