System Prompts Leaks

System Prompts Leaks is an open-source GitHub repository maintained by user asgeirtj that collects and publishes extracted system prompts from major AI products. Created in May 2025, the project has accumulated over 40,000 stars and 6,700 forks, and was featured in The Washington Post in May 2026. It is released under the MIT License and accepts community contributions via pull requests.

What It Is

This repository is a reference archive of raw system prompts — the hidden instructions that AI companies inject into their models before user conversations begin. It covers prompts from Anthropic (Claude), OpenAI (ChatGPT and Codex), Google (Gemini), xAI (Grok), Perplexity, GitHub Copilot, and a growing "Misc" category that includes tools like Cursor, Notion AI, Raycast AI, and Docker Gordon AI. The prompts are stored as Markdown files organized by vendor folder, making them easy to browse, compare, and reference.

What the Repository Contains

The collection spans dozens of models and product variants across major AI providers:

• Anthropic: Claude Opus 4.7, Opus 4.6, Sonnet 4.6, Claude Code, Claude Mobile (iOS), Claude for Excel/Word/PowerPoint, and older variants back to Sonnet 3.7
• OpenAI: GPT-5.5 (Thinking, API, Codex), GPT-5.4, GPT-5.3, o3, o4-mini, ChatGPT Atlas, tool-specific prompts (web search, deep research, canvas, memory), and image safety policies
• Google: Gemini 3.5 Flash, 3.1 Pro, 3 Flash, Gemini CLI, Antigravity CLI, Google Search AI Mode, Jules, NotebookLM
• xAI: Grok 4.3 Beta, 4.2, 4.1 Beta, Grok 4, Grok 3, personas, and safety instructions
• Misc: Cursor, Brave Search, Character AI, GitHub Copilot CLI, Meta AI, Warp 2.0 Agent, Zed AI, and others

How It Gets Updated

The repository is updated regularly — the README tracks a "Recently Updated" table showing the most recent additions with dates. As of late May 2026, the most recent additions include Perplexity Computer, VS Code Copilot Agent, Docker Gordon AI, and Gemini 3.5 Flash. Community members can contribute by submitting pull requests with raw prompt text as .md files in the appropriate vendor folder, or by opening issues to request specific models.

Why It Got Attention

The project was featured in The Washington Post on May 11, 2026, in an article titled "See the hidden rules behind AI. Then use them to rewrite this article." The repository's GitHub topics tag it under prompt-engineering, education, and open-source, reflecting its dual use as both a transparency resource and a learning tool for understanding how AI products are configured. The star count of over 40,000 and fork count of over 6,700 (as reported on the repository page) indicate significant community interest.

Update: Active as of May 2026

The repository was last pushed to on May 21, 2026, and continues to receive frequent updates. It was created on May 3, 2025, and has maintained active commit history throughout. The project direction is toward broader coverage — adding more models, more product integrations, and more variant prompts (e.g., personality modes, API vs. web variants) as they become extractable.