Ever wondered what instructions OpenAI, Anthropic, and Google secretly feed their AI models? This repository is a goldmine of extracted system prompts - the hidden directives that shape how ChatGPT, Claude, and Gemini behave. While these companies keep their prompt engineering under wraps, the community has been methodically extracting and documenting these internal messages through clever prompt injection techniques.
The collection reveals fascinating insights: how different companies handle safety guardrails, structure their AI personalities, and implement behavioral constraints. You’ll find everything from ChatGPT’s content policy instructions to Claude’s reasoning frameworks and Gemini’s response formatting rules. It’s like having the recipe book for modern AI assistants.
Whether you’re building your own AI application, researching prompt engineering techniques, or just curious about the wizardry behind the curtain, this repository offers unprecedented access to professional-grade system prompts. With 30k+ stars and active community contributions, it’s become the definitive reference for understanding how the world’s most sophisticated chatbots actually work under the hood.
⭐ Stars: 30150
💻 Language: JavaScript
🔗 Repository: asgeirtj/system_prompts_leaks