AI agents are becoming another runtime for enterprise data. They connect to APIs, files, databases and business systems, often with broad access and increasingly autonomous capabilities.

Security teams typically focus on identity, authentication and runtime protection. But once access succeeds, a different problem begins: what data is the agent actually consuming, how sensitive is it, what is it allowed to do with that information, and can we reconstruct those actions afterwards?

This session explores the security blind spot between AI agents and enterprise data. We will look at classification, policy enforcement, monitoring and investigation, using Microsoft Purview as one practical implementation example. We will also examine where controls lose visibility when agents interact across platforms, models and external services.

Attendees will leave with a practical way to evaluate agent architectures across four questions: What can the agent access? What data is involved? What can it do with it? And can you prove what happened afterwards?

Das gefällt mir für den Summit sogar besser als „Your AI Agent Has Access. Now Prove What It Did.“, weil es breiter ist und nicht so stark nach Audit Talk klingt.