People keep saying AI will replace humans, but only if it can figure out how to shut up about goblins. Starting with OpenAI’s ChatGPT‑5.1 model, users began to notice responses mentioning these mythical creatures a lot. So much so that a prompt had to be added to stop it from talking about them and several other real and imagined critters.
On April 27, 2026, a developer under the handle @arb8020 found a snippet from the OpenAI open source Codex GitHub repository, with a file named models.json. “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.”
He posted it to X, and the internet did its thing by making it into a meme. The most fantastical theory being that ChatGPT is really just run by goblins, explaining their data center’s massive water consumption.
Our theory, that ChatGPT is just really into Marvel villains, was promptly shot down.
Mo Goblins Mo Problems
By April 29th, OpenAI responded and explained how the whole goblin obsession came to be. The problem started within ChatGPT’s personality customization features, specifically the nerdy personality. The company acknowledges that in November, after the launch of 5.1, user complaints prompted an investigation into verbal tics. It showed that the “use of ‘goblin’ in ChatGPT had risen by 175% after the launch of GPT‑5.1, while ‘gremlin’ had risen by 52%.”
According to OpenAI the nerdy personality was told to be; “unapologetically nerdy, playful and wise AI mentor to a human. You are passionately enthusiastic about promoting truth, knowledge, philosophy, the scientific method, and critical thinking. […] You must undercut pretension through playful use of language. The world is complex and strange, and its strangeness must be acknowledged, analyzed, and enjoyed. Tackle weighty subjects without falling into the trap of self-seriousness.”
Because of this, the system was rewarded for using things like goblins in responses, causing it to use them more. Which might have been fine considering nerdy made up only 2.5% of all ChatGPT responses. However, it was responsible for 66.7% of all mentions of goblins, and it’s infatuation with them did not stay relegated. Soon other personalities were mentioning goblins due to the rewards and by 5.4 it was a widespread issue.
“The rewards were applied only in the Nerdy condition, but reinforcement learning does not guarantee that learned behaviors stay neatly scoped to the condition that produced them. Once a style tic is rewarded, later training can spread or reinforce it elsewhere, especially if those outputs are reused in supervised fine-tuning or preference data,” the company explains.
What’s Their Beef With Raccoons?
But as you may recall, the issue wasn’t just relegated to goblins; ChatGPT was also barred from discussing gremlins, raccoons, trolls, ogres, and pigeons. Apparently, these words also clustered with the original goblin issue. Though interestingly, the system seems to use frogs appropriately.
Despite the nerdy personality being discontinued in March, ChatGPT‑5.5 started training before the cause of the issue was found. And, of course, when developers were testing it, goblins abounded. Hence the need for the prompt barring them being mentioned.
This issue underscores an ever-increasing issue with AI. A simple and probably accidental preference during training caused a model to be overtaken by a goblin fascination. Between that and the near-constant battle against hallucinations, why are we still being told this will replace us?
Some things just need a human, non-goblin touch. But if you like your information overrun with mythical creatures, OpenAI included the prompt to get them back:
instructions=$(mktemp /tmp/gpt-5.5-instructions.XXXXXX) && \
jq -r '.models[] | select(.slug=="gpt-5.5") | .base_instructions' \
~/.codex/models_cache.json | \
grep -vi 'goblins' > "$instructions" && \
codex -m gpt-5.5 -c "model_instructions_file=\"$instructions\""
