Track the hidden system prompts behind major AI services
Have you ever wondered what rules operate behind the screen when you chat with AI services like ChatGPT or Claude? This repository publishes the system prompts of major AI services, updated daily since May 2025. It allows you to view the hidden instructions that AI follows without user input, exactly as they appear in the original text.
Recently Updated Prompts
As of November 3, 2026, the safety classifier prompt for Claude Code's automatic mode is the latest addition. At the end of October, prompts for Kagi Assistant, Perplexity Computer, MiniMax M3, LM Studio Bionic, GPT-Live-1.1 voice, Antigravity IDE, Grok Bot, Perplexity Search, Claude Motion, ZCode Desktop, and Cursor Cloud Agents were updated. Since specific instructions vary by model version and platform, this helps in understanding the behavior of the latest versions.
Included Services and Models
Service prompts from major AI companies such as Anthropic, OpenAI, Google, xAI, and Meta are organized here.
- Chat Interfaces Provides prompts for services that converse with general users, such as ChatGPT, Claude.ai, Gemini, and Grok, broken down by model version.
- Developer Tools Includes system prompts and sub-agent instructions for coding assistance tools like Claude Code, Codex, Cursor, and Copilot.
- Search and Browser Provides prompts for search and browser-based AI assistants such as Perplexity, Brave Search, and Comet Browser.
- Voice and Multimodal Also includes prompts related to voice AI and image generation, such as GPT-Live-1.1 and ElevenLabs.
Media and Expert Attention
This repository has been cited as a reference in articles from The Washington Post and CEPS's AI World. Simon Willison described it as the 'most thorough collection of leaked system prompts I know of.' It is used as a reliable source by researchers and journalists seeking to understand the internal workings of AI. Those studying prompt engineering also refer to it to analyze the structure of prompts used in real-world industry settings.
Things to Check Before Use
While this repository is based on publicly available information, it does not reflect the latest prompts for all AI services in real time. To check if a specific model's prompt has been updated, it is best to review the repository's recent change history. Additionally, this material is intended for learning and research purposes and may differ from the official documentation of each AI service. If you directly reuse the prompts, you must always check the terms of service of the respective service.
