As AI becomes embedded across identity platforms—from access recommendations to anomaly detection—many organizations are overlooking a critical reality: AI doesn’t fix broken identity programs. It learns from them.

This session examines how years of identity shortcuts—poor joiner data, undocumented roles, over-entitlement, and “temporary” access that never expired—are being absorbed into AI-driven decision engines. Once these patterns are learned, they don’t just persist; they scale.

Drawing from real-world enterprise identity deployments, this talk explores how identity AI models inherit organizational behavior, why traditional governance controls struggle to correct learned patterns, and how small upstream decisions can create outsized downstream risk once automation is introduced.

Attendees will gain a practical understanding of where AI systems are already learning from flawed identity data, why “we’ll fix the process later” fails once models are trained, and how to prevent identity automation from locking in past mistakes.