asgeirtj / system_prompts_leaks

2025-05-03 · Source: Github Trending: All languages · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cybersecurity & Data Privacy · Depth: Intermediate, medium

Summary

The "System Prompts Leaks" GitHub repository compiles and documents system prompt instructions for major AI chatbots, including Claude, ChatGPT, and Gemini. This resource provides insights into the underlying directives guiding these models' behaviors. Recent updates include prompts for GitHub Copilot for macOS (June 18, 2026), Claude Design (June 18, 2026), and GPT-5.5 Codex (June 18, 2026). It also features the Claude Fable 5 prompt from June 9, 2026, with a diff against Opus 4.8, and Gemini 3.5 Flash from May 20, 2026. The repository organizes prompts by vendors like Anthropic, OpenAI, Google, xAI, Perplexity, and Microsoft, covering various models, integrations, and tools, often obtained through prompt leakage techniques.

Key takeaway

For AI Engineers and Prompt Engineers developing or deploying large language models, understanding leaked system prompts is crucial for security and performance. You should regularly consult resources like this repository to identify potential prompt injection vulnerabilities and analyze how model updates, such as the Claude Fable 5 changes, impact core directives. This knowledge allows you to refine your own prompt strategies and harden AI applications against unintended disclosures.

Key insights

System prompts for major AI models are discoverable, revealing underlying operational directives and model behaviors.

Principles

• AI system prompts are not fully opaque.
• Prompt engineering can reveal model instructions.
• Model updates alter core system directives.

Method

System prompts are extracted by prompting AI chatbots to repeat their internal instructions, as demonstrated by ChatGPT and confirmed by Claude.

In practice

• Review leaked prompts for specific model behaviors.
• Compare prompt diffs to understand model evolution.
• Test prompt injection techniques on new models.

Topics

• System Prompts
• Prompt Engineering
• AI Security
• Large Language Models
• Claude
• ChatGPT
• Gemini

Code references

• asgeirtj/system_prompts_leaks

Best for: AI Architect, Machine Learning Engineer, NLP Engineer, Prompt Engineer, AI Engineer, AI Security Engineer

Related on AIssential

• Prompt Injection: The #1 AI Security Threat Every Developer Must Understand — Prompt injection is the top AI security threat, exploiting LLMs by overriding system instructions with malicious user input.
• Fictional Framing Part 2: Testing a Fix, Not Just Finding a Bug — Explicitly defining instruction scope, even for fictional contexts, can prevent large language model prompt leaks.
• Changes in the system prompt between Claude Opus 4.6 and 4.7 — Anthropic's Claude Opus 4.7 system prompt updates enhance safety, user experience, and tool integration.
• The Playbook for System Prompting — System prompting is a systems intervention, not just instruction, requiring translation of tacit knowledge into effective conditioning signals.
• More Prompts = Worse Code? — Complex or outdated prompts create technical debt, silently degrading LLM performance due to model-specific decay.
• Attackers prompted Gemini over 100,000 times while trying to clone it, Google says — Model extraction via extensive prompting is a common method for AI model distillation and intellectual property theft.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Github Trending: All languages.