wtf is leaked-llm-prompts?

This project is a collection of "leaked" prompts, the hidden instructions that AI companies build into their chatbots behind the scenes. If you've ever wondered what ChatGPT or Claude is told to do before you even type your first message, this repo gathers examples of those instructions in one place for anyone to browse. Most AI assistants don't just wing it when you talk to them. Under the hood, they receive a detailed prompt from their creator, things like "you are a helpful assistant," "refuse certain types of requests," "format your answers this way," or "you do not have emotions." These system prompts shape the AI's personality, boundaries, and behavior. The prompts in this collection were reportedly extracted or reverse-engineered from various AI products, giving you a peek at the guardrails and instructions that companies put in place. This would be useful for a few types of people. A founder building an AI product might study these to see how major companies structure their prompts and handle edge cases. A prompt engineer or curious tinkerer could learn techniques for writing better instructions by seeing how the big players do it. And anyone interested in AI transparency might simply want to understand what's going on under the hood of tools they use every. The README doesn't go into detail about where each prompt came from, how they were obtained, or which specific AI products they belong to. It's essentially a raw collection with no framing, commentary, or usage guidance, so you'd need to open the files and judge for yourself what you're looking at and how credible each example is.