Louise is a senior penetration tester and operations partner at Xanthus Security, where she’s spent the last six years breaking into everything from critical infrastructure to mobile health apps. Her work focuses on real-world application security, uncovering logic flaws, testing assumptions, and finding out what actually breaks when things get weird.
Lately, that means testing AI chatbot integrations across web and mobile apps, and helping teams understand where the trust boundaries really are. She’s passionate about sharing practical techniques that other hackers and builders can immediately use in the field. You’ll probably find her in the CTF room grinning at chaos and cheering people on.
Louise Powell
All Sessions by Louise Powell
-
Day 1 - Thursday
-
11:00
Jailbreak Me: Hacking the Brain of AI Chatbots
11:00 - 11:45For years, we tested forms. Then we tested APIs. Now we're testing the brains inside your apps. In this fast-paced, practical talk, we'll walk through the emerging art of hacking AI chatbots from the perspective of a penetration tester. You'll see how prompt injection works (and the many forms it can take), why indirect manipulation is especially dangerous, and what developers get wrong when embedding AI assistants into apps, especially when those assistants act as logic engines, personal coaches, or API middlemen. We'll break down the anatomy of a typical chatbot-powered feature, explore real-world examples, and walk step-by-step through how to bypass filters, extract sensitive data, and hijack intent. As automated tools and AI begin to catch traditional bugs, this emerging attack surface demands a new skillset. We'll see how chatbot security isn't just about filtering obvious words, it's about detecting intent, even when that intent is buried or disguised. If you're testing modern apps and ignoring the chatbot, you're leaving the brain unguarded.