(PI36) (INCYBER) Automating AI agents security and evaluation at enterprise scale

As AI agents move from pilot to production, manual testing and red teaming is no longer scalable. How do you continuously validate the security of your AI when the underlying models and data are constantly changing?

This session provides an approach for securing conversational AI agents through automated continuous red teaming. We will break down the anatomy of AI agent vulnerabilities (data leakage, hallucinations, and logic bypasses) and show how to automate the detection of these risks. Learn how to ground your AI evaluations in your company’s unique context using Knowledge Bases, and how to implement multi-turn persona simulations that mimic real-world user behaviors.