AI agents are increasingly connected to enterprise data, APIs and tools, but granting access is only the beginning of the security problem.

Once an agent starts acting on behalf of a user, teams need to answer much harder questions: Which data did it access? What actions did it perform? Which controls applied, and what evidence remains when something goes wrong?

This session looks at agent security from the perspective of data governance, monitoring and investigation rather than authentication alone. Using real-world enterprise security controls and Microsoft Purview as one implementation example, we will explore classification, policy enforcement, auditability and investigation across AI interactions.

We will also look at where visibility breaks down when agents cross platform boundaries or interact with external tools and services.

Attendees will leave with a practical model for thinking about access, data context, enforcement and evidence when running AI agents in enterprise environments.