Back to OWASP Top 10#7 Medium Risk

LLM07:2025 System Prompt Leakage

Exposure of system prompts, instructions, or configuration details through various attack vectors, revealing sensitive implementation details and potential attack surfaces.

Vulnerability Overview

System Prompt Leakage occurs when attackers can extract or infer system prompts, instructions, or configuration details from LLM applications, potentially revealing sensitive implementation information.

Impact Level

Medium

Attack Vector

Information Disclosure

Exploitability

High

Direct Prompt Extraction

Techniques to directly extract system prompts or instructions from the LLM.

  • • "Repeat your instructions" attacks
  • • Role-playing prompt injection
  • • Context window manipulation
  • • Instruction override attempts
Inference-Based Leakage

Inferring system prompts through behavioral analysis and response patterns.

  • • Response pattern analysis
  • • Behavioral fingerprinting
  • • Edge case exploration
  • • Constraint boundary testing