Abstract
A new take on our secure AI agents workshop. Instead of building from scratch, we start with an existing Quarkus LangChain4j application and try to break it, then fix it. Working through a series of real attacks, you'll exploit and then defend against prompt injection, broken authorization, excessive agency, and sensitive information disclosure. Along the way you'll learn how to carry a user's identity all the way to where each decision is made and apply defense in depth to AI agents using Quarkus and LangChain4j.