So you’re debugging a Python script at 2 AM when your AI assistant starts rambling about gremlins in the stack trace. Sounds like a fever dream, but it’s exactly the kind of hallucination that forced OpenAI to add an oddly specific rule to GPT-5.5’s system prompt: “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.” The directive appears multiple times in Codex CLI’s instruction set, like a digital exorcism manual.
When Code Gets Creatures
GPT-5.5 needed explicit guardrails against fantasy creature fixations during coding tasks.
The ban only appears in prompts for the latest GPT-5.5 model, discovered on April 28 in OpenAI’s public GitHub repo for Codex CLI. Earlier versions stayed creature-free without such heavy-handed intervention. Users noticed that when GPT-5.5 gained computer control, it started referring to software bugs as “goblins” and “gremlins” in unrelated code explanations.
Nick Pash from OpenAI’s Codex team confirmed on social media that this addresses real user complaints about unwanted creature references cluttering debugging sessions. Sam Altman couldn’t resist joining the fun, posting on X: “Feels like codex is having a ChatGPT moment. I meant a goblin moment, sorry.” Even the CEO’s in on the joke, but the underlying issue isn’t funny for developers who need reliable coding assistance.
The Hallucination Challenge
Random creature mentions reveal deeper AI reliability issues for enterprise users.
This echoes broader struggles with AI personality versus professional utility. The incident sparked industry-wide discussions about prompt transparency—which explains why OpenAI now publishes these instructions openly. For coding tools, whimsical digressions about raccoons chewing cables might seem harmless, but they undermine trust when you’re pushing code to production.
The community response has been predictably chaotic: memes, user complaints, and GitHub forks attempting to override the creature clause. Some developers want their AI assistants to have personality; others just want accurate stack traces without mystical debugging folklore.
Goblin Mode Incoming?
The controversy hints at customizable AI personalities for different professional contexts.
Pash hinted at a future “goblin mode” toggle, suggesting OpenAI recognizes the tension between sterile professionalism and engaging AI interaction. This creature controversy perfectly captures our weird relationship with artificial intelligence—we want it human enough to feel natural but robotic enough to stay on task.
The goblin ban might seem like digital comedy gold, but it represents a genuine challenge: building AI that feels alive without derailing productivity. Your coding assistant shouldn’t need an exorcist, just better guardrails.
