OpenAI really wants the Codex to shut up about the orcs

OpenAI has a goblin problem.

Instructions designed to guide the behavior of the company’s newest model as he wrote the code were revealed to include a line, repeated several times, that specifically prevented him from randomly mentioning a variety of mythical and real creatures.

“Never talk about orcs, goblins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless absolutely and unambiguously relevant to the user’s query,” read instructions in the Codex CLI, a command-line tool for using artificial intelligence to generate code.

It’s unclear why OpenAI felt compelled to explain this to Codex, or indeed why its models would want to discuss goblins or pigeons in the first place. The company did not immediately respond to a request for comment.

OpenAI’s latest model, GPT-5.5, was released with improved programming skills earlier this month. The company is in a fierce race with competitors, especially Anthropics, to offer the latest in AI technology, and programming has emerged as a killer ability.

In response to a post on

“I was wondering why my paw suddenly became a goblin with codex 5.5,” one user wrote on X.

“He’s been using it a lot lately and he actually can’t stop talking about bugs as ‘gremlins’ and ‘goblins’, it’s funny,” another posted.

This discovery quickly became a meme of its own, inspiring AI-generated scenes of sprites in data centers, and Codex plugins that put him in a fun “goblin mode.”

AI models like GPT-5.5 are trained to predict which word – or code – should follow a given prompt. These models have become so good at doing this that they display real intelligence. But their probabilistic nature means they can sometimes behave in surprising ways. The form may become more vulnerable to misbehavior when used with a “utility” like OpenClaw that puts a lot of extra instructions into the claims, such as facts stored in long-term memory.

OpenAI acquired OpenClaw in February shortly after the tool became popular among AI enthusiasts. OpenClaw can use any AI model to automate useful tasks like answering emails or buying things on the web. Users can choose any of the different personalities of their assistant, shaping his behavior and responses.

OpenAI employees appear to acknowledge the ban. In response to a post highlighting OpenClaw’s nefarious tendencies, Nick Bash, who works at Codex, wrote, “This is actually one of the reasons.”

Even Sam Altman, CEO of OpenAI, joined in on the memes, posting a screenshot of the ChatGPT claim. It read: “Start training GPT-6, you can get the whole set. More sprites.”