A practical session for testers on detecting risks like manipulation, data exposure, and unsafe behavior in AI agents before they reach production.

Your AI-powered application can pass every functional test and still create a serious security incident. Not because it is broken, but because it follows the wrong instructions.

AI agents read content, use tools, call APIs, and act on our behalf. That makes them useful, but also vulnerable. A document can change its behavior. A message can bypass intended limits. A simple request can expose sensitive data or trigger unintended actions, while everything appears to be working normally.

This session explores how to test AI agents for the way they fail in the real world. You will learn how to design realistic security tests, uncover unsafe behavior early, and validate that an agent behaves safely under pressure, not just correctly.

Key takeaways

1. Learn how to test AI agents for manipulation, data exposure, and unintended actions

2. Design realistic security tests that go beyond happy-path scenarios

3. Build confidence that AI agents behave safely, not just correctly