OpenAI Codex CLI System Prompt Leaked: 'Never Talk About Goblins' and Bizarre Content Restrictions Exposed

Ground News·medium signal

The GPT-5.5 Codex CLI system prompt was extracted and published on April 26, revealing internal base_instructions that include '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.' The leak was catalogued in a GitHub repository alongside prompts from Claude Code, Gemini CLI, and Grok. OpenAI subsequently patched leaked review prompts from Codex TUI transcripts. Simon Willison flagged the finding on April 28.