jujumilk3/leaked-system-prompts

This repository is a public archive of leaked system prompts from popular services that are built on top of large language models. A system prompt is the hidden instruction that a company gives to a chatbot or AI assistant before any user message arrives. It sets the assistant's persona, rules, and limits. Companies usually keep these prompts private, but users sometimes get the model to reveal them through clever questioning, and this repo collects those revealed prompts in one place. The README itself is short and to the point. It describes the project in two sentences and then lays out the contribution rules. The maintainer asks contributors to follow the same document format as existing entries and to include sources that can be verified, or prompts that can be reproduced by anyone else. This is the project's way of keeping the collection honest, since a prompt with no source is just a claim. For people who do not want to open a full pull request, there is a lighter option. You can post a link in the Issues section instead. If the maintainer can verify the source or reproduce the prompt, they will merge it into the main collection themselves. This lowers the barrier for casual contributors while keeping a verification step in place. The README closes with a legal note. The repository is cited in academic papers, so the maintainer wants to keep it online and avoid DMCA takedown requests from companies. Contributors are asked not to include sensitive commercial source code, only the prompt text itself. That is the entire scope of the project: a curated, source-checked, citation-friendly archive of system prompts that have escaped into public view.