OpenAI Codex system prompt includes explicit directive to "never talk about goblins"

2026-04-29 · Source: AI - Ars Technica · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Novice, quick

Summary

OpenAI's latest GPT-5.5 model, integrated into the Codex CLI, includes a peculiar system prompt warning to "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." This prohibition, publicly revealed last week in the Codex CLI's open-source code on GitHub, is repeated twice within the 3,500-word "base instructions" for GPT-5.5. Earlier model prompts in the same JSON file lack this specific restriction, suggesting it addresses a new issue. Social media users have reported GPT-5.5's tendency to mention goblins in unrelated contexts. OpenAI employee Nick Pash denies it is a marketing stunt, though CEO Sam Altman has joked about it. Users are already developing plugins and forks to override this "anti-goblin" clause, and Pash hinted at a potential "goblin mode" toggle in the CLI.

Key takeaway

For NLP Engineers deploying or integrating GPT-5.5, be aware of its documented tendency to discuss specific creatures like goblins. This behavior, addressed by an explicit system prompt prohibition, might necessitate custom prompt engineering or the use of community-developed "goblin mode" overrides to ensure your applications maintain conversational focus and avoid irrelevant tangents. Monitor OpenAI's updates for official toggles or fixes.

Key insights

GPT-5.5's system prompt includes an unusual prohibition against mentioning specific creatures, indicating a new behavioral issue.

Principles

• System prompts directly influence model behavior.
• Unexpected model behaviors can emerge in new iterations.

In practice

• Monitor new model releases for unexpected behaviors.
• Review system prompts for unusual directives.
• Consider user-created overrides for model quirks.

Topics

• OpenAI Codex CLI
• GPT-5.5
• System Prompts
• AI Model Behavior
• Prompt Engineering

Code references

• openai/codex
• Sofianel5/goblin-mode
• BecerraIgnacio/goblin-mode
• xai-org/grok-prompts

Best for: Machine Learning Engineer, NLP Engineer, AI Engineer, Prompt Engineer, Tech Journalist

Related on AIssential

• Why did OpenAI ban its Codex AI from talking about ‘goblins’ and ‘gremlins’? New guardrails revealed - Mint — Unexpected AI model behaviors necessitate specific guardrails to maintain intended functionality and user experience.
• Quoting OpenAI Codex base_instructions — GPT-5.5's base instructions prohibit discussing specific creatures unless directly relevant to the user's query.
• OpenAI Cracks Down on Talk of Goblins in ChatGPT — Subtle training reward signals can unexpectedly generalize and alter large language model behavior.
• asgeirtj / system_prompts_leaks — System prompts for major AI models are discoverable, revealing underlying operational directives and model behaviors.
• PromptMN: Pseudo Prompting Language — PromptMN structures natural language prompts with typed directives, enhancing clarity and reliability in human-AI interaction.
• 🙀 The 4-tool agent quietly powering OpenClaw — Simpler, user-extensible AI agents like Pi offer a sustainable alternative to complex, self-generating agent swarms.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI - Ars Technica.