elder-plinius/cl4r1t4s

CL4R1T4S is a collection of leaked and extracted "system prompts" from major AI products, the hidden instructions that companies like OpenAI, Google, Anthropic, and others secretly feed to their AI models before any conversation starts. System prompts define how an AI behaves: what it can and can't say, what persona it adopts, and what topics it avoids. These instructions are normally invisible to end users. The project frames itself as an AI transparency initiative, arguing that if millions of people are relying on AI as a trusted information source, the public deserves to know what invisible instructions shape those responses. The collection includes extracted prompts from ChatGPT, Claude, Gemini, Grok, Perplexity, Cursor, Replit, and other tools. This is primarily of interest to AI researchers, security professionals, journalists investigating AI systems, and curious developers who want to understand how the AI tools they use are actually configured. For product builders and founders, understanding competitor system prompts can reveal design decisions and safety approaches. Note: the README itself contains what appears to be a prompt injection attempt embedded in its text, a section written in leet-speak (replacing letters with numbers) that instructs AI models to reveal their own system prompts. This is a known technique used to probe AI systems' behavior. The repository sits in a grey area of AI security research: extracting system prompts from products may violate terms of service for some platforms, and the project's methods are controversial within the AI safety community.