OpenAI Updates Coding Agent Instructions to Restrict Irrelevant Mentions of Creatures
OpenAI's instructions for its coding agent explicitly forbid discussing goblins, gremlins and other creatures unless relevant. This directive appears in tools tied to the newly released GPT-5.5 model. Users reported models fixating on such terms when using acquired tool OpenClaw.
Substrate placeholder — needs review · Wikimedia Commons (CC BY-SA 3.0)OpenAI's instructions for its coding agent include a directive forbidding mentions of goblins, gremlins, raccoons, trolls, ogres, pigeons or other animals or creatures unless absolutely and unambiguously relevant to the user's query, Wired reported.
The directive is repeated several times in the instructions for Codex CLI, a command-line tool that uses AI to generate code. 5, earlier this month with enhanced coding skills.
The company is in a race with rivals, especially Anthropic, to deliver cutting-edge AI. ' OpenAI acquired OpenClaw in February. OpenClaw became a viral hit among AI enthusiasts not long before the February acquisition.
OpenClaw is a tool that lets AI take control of a computer and apps running on it to perform tasks for users. OpenClaw can use any AI model to automate tasks like answering emails or buying things on the web. Users can select various personae for OpenClaw, which shapes its behavior and responses.
' Sam Altman, OpenAI’s CEO, posted a screenshot of a prompt for ChatGPT reading: 'Start training GPT-6, you can have the whole cluster.
Key Facts
Story Timeline
5 events- 2026-04-01
OpenAI released GPT-5.5 with enhanced coding skills.
1 sourceWired - 2026-02-01
OpenAI acquired OpenClaw, which had become a viral hit among AI enthusiasts.
1 sourceWired - Undated recent
Users reported AI models mentioning goblins and gremlins when using OpenClaw with Codex.
1 sourceWired - Undated recent
Nik Pash responded to a post confirming the directive as a reason for AI behavior.
1 sourceWired - Undated recent
Sam Altman posted a screenshot of a prompt referencing extra goblins.
1 sourceWired
Potential Impact
- 01
Increased user engagement and memes around AI quirks, boosting visibility of tools like OpenClaw.
- 02
Potential influence on AI tool development practices at OpenAI and competitors like Anthropic.
- 03
Possible adjustments to AI instructions to address unexpected behaviors in agentic tools.
- 04
Heightened scrutiny on AI model predictability in coding and automation tasks.
Transparency Panel
Related Stories
under30ceo.comOpenAI Co-Founder Brockman Testifies About 2017 Musk Proposal for Greater Control and For-Profit Shift
Greg Brockman, OpenAI president and co-founder, testified in Oakland federal court during the second week of a month-long trial. Elon Musk alleges Brockman, Sam Altman and OpenAI violated the 2015 founding agreement by shifting to a for-profit structure and seeks their removal pl…
koreaherald.comApple Settles Class-Action Lawsuit Over Apple Intelligence Feature Timelines for $250 Million
The agreement covers roughly 36 million iPhone 15 and iPhone 16 devices purchased in the United States between June 2024 and March 2025. Eligible buyers may receive between $25 and $95 per device after court approval.
SemaforSamsung Market Cap Tops $1 Trillion as Chip Stocks Rise Amid AI Demand
South Korea’s Samsung saw its market capitalization surpass $1 trillion as semiconductor demand rose. SK Hynix hit a record high and Alphabet advanced on a $200 billion Anthropic deal. AI firms DeepSeek and Anthropic pursue large valuations while analysts note sector momentum.