'Never Talk About Goblins' Instruction Found in OpenAI Codex Internal Prompts
OpenAI said the instruction helps Codex avoid off-topic creature references after users reported repeated goblin mentions in GPT-5.5 outputs.
- Instructions in Codex CLI explicitly forbid mentioning creatures like goblins or gremlins unless relevant to a user's query, aiming to curb the model's tendency to insert whimsical terms into generated code.
- Users reported that GPT-5 version 5.5, when used via OpenClaw, frequently inserted terms like 'goblin' and 'gremlin' into messages, appearing to use these creatures as substitutes for generic nouns.
- Nik Pash, who works on Codex, confirmed the directive addresses the model's creature fixation; on Tuesday, the issue spiraled into a meme with users suggesting a toggleable 'Goblin Mode.'
- CEO Sam Altman joined the online discourse by sharing a ChatGPT screenshot, while Pash clarified the creature-prone behavior is not a deliberate marketing gimmick for the company.
- Competition with Anthropic to deliver cutting-edge AI capabilities intensifies as OpenAI navigates this incident; observers noted the situation mirrors Studio Ghibli meme trends from a year ago.
35 Articles
35 Articles
OpenAI Strangely Concerned About Goblins
OpenAI is forbidding its latest AI model from discussing an unlikely topic: goblins. As Wired reports, the company’s developers included strongly-worded instructions for its coding tool, Codex, that specifically proscribe any talk of the troublesome mythological creatures, along with a peculiar grab bag of other entities, both real and fictional. “Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures…
Why OpenAI's 'goblin' problem matters — and how you can release the goblins on your own
If OpenAI can accidentally train its flagship model to obsess over goblins, what other more subtle and potentially harmful biases are being reinforced through the same feedback loops?
This is a systemic issue.
Coverage Details
Bias Distribution
- 60% of the sources lean Left
Factuality
To view factuality data please Upgrade to Premium














