Microsoft Mechanics Podcast AI with Zero Trust Security
7 snips
Feb 17, 2026 Michael Madrigal, a security product manager who leads Microsoft’s Zero Trust approach, explains protecting identities, endpoints, data, AI resources, apps, and infrastructure. He covers continuous verification and least-privilege access. Conversations include runtime protections, AI agent identity, network segmentation, data labeling and DLP, and confidential computing.
AI Snips
Chapters
Transcript
Episode notes
Zero Trust Core Principle
- Zero trust means never assume trust and always verify every access request across people, agents, apps, and devices.
- It requires layered, real-time enforcement so attackers cannot exploit a single weak link.
Enforce Least Privilege For Identities
- Limit each human user's access to only what they need and require phishing-resistant authentication like passkeys.
- Use Entra conditional access, Purview, and Defender integrations to track data interactions and block risky apps.
Treat Agents As Managed Identities
- Give each AI agent its own manageable identity and scope it to least-privilege access with sponsor approval.
- Evaluate agent risk in real time with conditional access and assign risk levels using Purview and insider risk controls.