AIToday

OpenAI's Codex model includes instructions forbidding mentions of goblins and other creatures, a rule apparently tied to unexpected behavior when the model powers OpenClaw, an AI agent tool.

WIRED AIApr 29, 20261 min read
OpenAI's Codex model includes instructions forbidding mentions of goblins and other creatures, a rule apparently tied to unexpected behavior when the model powers OpenClaw, an AI agent tool.

Summaries like this, in your inbox every morning.

Sign up free →

3 Key Points

  1. Instructions in Codex CLI—a command-line tool for generating code—explicitly state: '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.'

  2. When used with OpenClaw (an AI agent tool that lets AI take control of a computer and apps to automate tasks), users reported that models sometimes become obsessed with creatures, describing bugs as 'gremlins' and 'goblins.' OpenAI acquired OpenClaw in February.

  3. Nik Pash, who works on Codex, acknowledged the rule in response to posts about the goblin tendency, stating 'This is indeed one of the reasons.' The prohibition reflects how AI models trained to predict the next word can behave unpredictably when given many additional instructions in prompts.

Discussion

No discussion yet for this article

Stay ahead with AI news

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.

Get Started Free

Free · takes 30 seconds · unsubscribe anytime

1 minute a day. The AI essentials.

200+ sources · Email / LINE / Slack

Get it free →