System Prompts Leaks Repository Trends on GitHub
The GitHub repository `asgeirtj/system_prompts_leaks` gained over 5,100 stars in a month by compiling extracted system prompts from major commercial AI models. The collection documents secret baseline instructions from leading systems including Anthropic Claude, OpenAI ChatGPT, Google Gemini, xAI Grok, and Cursor. These extracted prompts give developers and prompt engineers rare practical insights into how top AI companies structure safety guardrails, format responses, and enable tool integration. It also underscores the security challenges of prompt leaking in modern large language model deployments. Maintained primarily in Python, the project has accumulated over 11,000 forks and receives regular updates contributed by the community. The repository spans prompt architectures across prominent platforms including OpenAI, Anthropic, Google, xAI, Cursor, and Kimi.
## BACKGROUND
A system prompt is a predefined hidden instruction provided to an AI model before any user interaction, setting its rules, tone, available tools, and safety boundaries. Prompt leaking refers to technique-driven extraction where users trick the model into revealing these hidden configuration instructions.