SAINTCON 2025 - Jake Williams - Findings From Real-World AI Application Assessments

Video Overview & Insights

Findings From Real-World AI Application Assessments

After performing numerous risk assessments and penetration tests of AI-enabled applications, we've noticed some significant patterns we're actually seeing in the wild. In this talk, Jake will discuss the methodology he uses for AI risk assessments. Then we'll discuss the common failures, including lack of observability into agent behavior, failing to measure bias impact in line of business apps, and more. We'll discuss specifically how those risks were uncovered. Finally, while nothing will completely extinguish this dumpster fire, we'll discuss practical controls that can at least keep the flames burning at a manageable level.

Main Track 2 (Ballroom B)