Securing AI Agents: Blueprints and Agent 365 with Derk van der Woude
Show notes
How do you secure AI agents in Microsoft Entra? Security MVP Derk van der Woude explains agent identity blueprints, classic vs modern agents and Agent 365.
In the relaunch of MVP Security Insights, Frans and Pouyan talk with Derk about AI agent security from both the attacker and the defender side.
In this episode
• What a secure AI assessment and a secure AI landing zone look like in practice
• Workload identity risk: multi-tenant apps with high-privileged application permissions
• Classic vs modern agents, and why most security controls only work on modern agents
• Agent identity blueprints: the four-layer model and why attackers go after the blueprint
• What you won't see in the UI: inherited permissions that only show up in the token
• Agent 365 explained: control plane, license and security controls in Entra, Defender, Purview and Intune
• Conditional Access for agents and the new AgentsInfo hunting table
• Where to start if you have one week to secure the agents in your tenant
Chapters
00:00 Welcome to MVP Security Insights
02:27 From a 486 to Microsoft Certified Master
07:12 Building a secure AI landing zone
11:57 Workload identity posture management
15:04 Classic vs modern agents
19:53 Agent identity blueprints explained
26:46 What the UI and audit logs miss
31:12 Agent 365 and Conditional Access
37:53 Where to start in one week
41:59 Rapid fire, IT/OT and advice
Guest: Derk van der Woude – CTO Cybersecurity at BBTG, Microsoft Security MVP (https://blue16.nl)
Hosts: Frans Oudendorp & Pouyan Khabazi
Watch on YouTube: https://youtu.be/JgH00RYLbHg
More episodes: https://talkingsecurity.nl