As AI agents gain access to enterprise data, security becomes the top priority. This workshop addresses the critical challenges of Prompt Injection and Sensitive Data Disclosure. You will implement a "defense-in-depth" strategy using Google Cloud Model Armor to filter malicious inputs and protect against data exfiltration. Learn how to combine reasoning-based defenses with deterministic controls like Agent Identity and IAM policies to ensure your agents are helpful, compliant, and most importantly unhackable.