asgeirtj / system_prompts_leaks

项目摘要

A comprehensive, regularly-updated collection of verbatim system prompts from major AI chatbots including Claude, ChatGPT, Gemini, Grok, Cursor, and others. Serves as a research and reference resource for understanding how commercial AI systems are configured and behave.

为什么值得关注

This is the most thorough collection of leaked system prompts available, cited by major publications like The Washington Post and recognized by prominent AI researchers like Simon Willison. It provides unprecedented transparency into how leading AI systems are instructed and constrained.

优势

• Unparalleled comprehensiveness across all major AI providers
• Regularly updated with new prompts and versions
• Validated and cited by major media outlets and researchers
• Open source with CC0 license for unrestricted use
• Community-driven with contributions from multiple researchers

局限性

• No automated validation or testing of prompt accuracy
• Relies on manual extraction which may introduce errors or be incomplete
• Legal and ethical gray area around collecting proprietary system prompts
• No formal versioning or release management

使用场景

• Research on AI alignment, safety, and system behavior
• Prompt engineering and understanding commercial AI configurations
• Comparative analysis of how different AI providers instruct their models
• Educational resource for learning about AI system design
• Journalistic investigation into AI system capabilities and constraints