Defending LLMs against Prompt Injection

Prompt injection attacks exploit the fact that Large Language Models treat instructions and user inputs identically. Attackers craft inputs that override system prompts, executing unauthorized actions.

Mitigation Strategies

• Strict prompt separation: Use clear delimiters to demarcate user inputs.
• Output validation filters: Scan LLM outputs for system prompt keywords.
• Least privilege permissions: Do not give LLM agents administrative tool execution rights.