explaingit

asgeirtj/system_prompts_leaks

📈 Trending★ 40,415Audience · researcherComplexity · 1/5ActiveLicenseSetup · easy

TLDR

A curated collection of extracted system prompts from ChatGPT, Claude, Gemini, Grok, and other AI assistants, the hidden instructions that shape how these models behave.

Mindmap

mindmap
  root((repo))
    What it does
      Collects system prompts
      From multiple AI models
      Organized by company
    Content included
      ChatGPT variants
      Claude models
      Gemini and Grok
      Perplexity prompts
    Use cases
      Study AI behavior
      Understand constraints
      Compare model instructions
      Research transparency
    Audience
      AI researchers
      Developers
      Policy makers
      Curious learners

Things people build with this

USE CASE 1

Study how AI companies instruct and constrain their models to understand baseline behavior and safety guardrails.

USE CASE 2

Compare system prompts across different AI assistants to see how companies approach persona, capabilities, and ethical guidelines differently.

USE CASE 3

Research AI transparency and accountability by examining what hidden instructions shape widely-used chatbots.

USE CASE 4

Understand why a specific AI model behaves the way it does by reading its actual system prompt.

Getting it running

Difficulty · easy Time to first run · 5min
Use freely for any purpose including commercial, as long as you keep the copyright notice.

In plain English

This repository is a collection of extracted system prompts from popular AI chatbots and coding assistants. A system prompt is a set of hidden instructions that AI companies write to shape how their models behave, what persona they adopt, what they will or will not do, and what tools they have access to. End users typically never see these instructions, but researchers have found ways to extract or reconstruct them by probing the models with carefully crafted questions. The collection includes system prompts from ChatGPT at various GPT versions, multiple Claude models from Anthropic including Opus and Sonnet variants as well as Claude Code, Google's Gemini family, xAI's Grok, and Perplexity. Each model's prompt is stored as a Markdown file organized by company, making them easy to browse and compare. You would find this repository useful if you are an AI researcher studying how these systems are instructed and constrained, a developer building on top of these APIs who wants to understand the baseline behavior you are working with or competing against, or a curious person who wants to understand why a chatbot behaves the way it does. It is also valuable for journalists and policy researchers studying AI transparency, since it reveals what behavioral guardrails and capabilities companies have programmed into their most widely used products. The project itself does not contain any code in the traditional sense. It is entirely a documentation and research archive of Markdown text files organized into folders by company. There is no programming language, framework, or runtime involved. The repository is updated regularly as new model versions are released, and pull requests are accepted to add new prompts.

Copy-paste prompts

Prompt 1
Show me the system prompt for ChatGPT GPT-4 and explain what behavioral constraints it includes.
Prompt 2
Compare the system prompts from Claude Opus and Claude Sonnet, what are the key differences in how they're instructed?
Prompt 3
What tools and capabilities does the Gemini system prompt grant the model, and how does that differ from ChatGPT?
Prompt 4
Use these extracted system prompts to identify common patterns in how AI companies handle safety and ethical guidelines.
Prompt 5
What does the Grok system prompt reveal about xAI's approach to content moderation compared to other AI assistants?
Open on GitHub → Explain another repo

Generated 2026-05-18 · Model: sonnet-4-6 · Verify against the repo before relying on details.