Prompt Injection

Prompt injection is a class of attack where adversarial text inside a user query, tool result, or external document overrides the AI system's instructions. Defenses include strict input/output sanitization, separation of trusted and untrusted text, and structured tool boundaries.