AI agents can access enterprise data, call tools and generate outputs across multiple systems, creating new paths for sensitive information to leave controlled boundaries.

This session follows the data flow through agent interactions and shows where traditional security controls still work, where visibility starts to break down, and where new risks emerge. Using Microsoft Purview as one practical example, we will look at classification, DLP, monitoring and investigation around AI-driven data access.

Attendees will leave with a practical way to identify and assess data exfiltration risks in agentic AI architectures.