BRK340 Breakout Thu 19 Nov 17:15 PST
A Zero Trust approach to securing AI runtime with Microsoft Defender
James Brotsos, Krishna Sunkavalli
Summary
From the recording / Microsoft AI summary
The session applies Zero Trust to AI agents at runtime in three steps: discover and inventory agents, improve their security setup, and protect them while they run. The Agent 365 registry and Defender inventory give one view across Copilot Studio, Foundry, Microsoft 365, third-party platforms and local devices, including shadow agents and MCP servers. Each agent gets a combined risk score. In a fictional demo, a Foundry procurement agent read a vendor PDF with hidden text and recommended a fabricated vendor; real-time protection blocked it and raised a Defender XDR alert. A Zero Trust workshop and trial licence were stated as available. Based on an AI-generated summary; no statuses or SKUs given.
Session description
Microsoft's catalogue description
AI agents are the next attack surface. Learn how Microsoft Defender helps apply Zero Trust, reduce agent risk, and defend against AI attacks such as prompt injection, tool misuse, and identity abuse, with guidance mapped to OWASP and MITRE.
Demo timestamps
Microsoft's player does not support links to a specific time, so open the recording and skip to the time shown.
- Skip to 8:20
Defender runtime protection for AI agents
Walkthrough of Microsoft Defender runtime protection for AI agents. Captions mark transitions at 8:20, 18:59–19:43, 26:48 and 32:08.