Back to glossary
AI GLOSSARY
Prompt Leaking
Security & Adversarial AI
A type of attack or unintended behavior where the contents of a confidential system prompt are revealed to end users, either through direct questioning, clever prompt engineering, or model vulnerabilities. Since system prompts often contain proprietary business logic, sensitive instructions, or security-relevant constraints, leaking them can expose intellectual property and undermine application security.