Prompt Injection Prevention

Defending against instructions hidden in untrusted input that hijack an AI agent's behavior.