Leaked-system-prompts is a collection repository that archives system prompts extracted from widely used LLM-based services.
The repository addresses the interest in understanding how commercial LLM systems are instructed and constrained at the prompt level. It works by gathering system prompts that have been discovered or disclosed from various AI services, organizing them in a standardized format, and making them publicly accessible for research and analysis. The project accepts contributions through pull requests with verified sources or reproducible prompts, or through issue submissions with links to verifiable evidence.
This repository is most useful for researchers, security professionals, and developers studying LLM behavior, prompt engineering techniques, and the design patterns used by commercial AI systems. It suits academic work, comparative analysis of different AI services' instruction strategies, and understanding potential attack surfaces or jailbreak vectors. The project explicitly avoids including sensitive commercial source code to mitigate legal risk, focusing instead on the system prompts themselves.
The project maintains a structured contribution process that requires verification of sources or reproducibility of prompts before merging. The maintainer actively reviews submissions and processes both formal pull requests and informal issue-based contributions. The repository has achieved sufficient academic recognition to warrant deliberate legal precautions in its guidelines, indicating sustained scholarly interest in its contents.
