Zero Trust security for AI agents
Apply Zero Trust controls to every AI agent in your environment across identity, tool usage, and data access. Extend Conditional Access in Microsoft Entra to evaluate every agent authorization request in real time against the same risk signals as human users. Assign each agent its own managed identity with Entra Agent ID and scope permissions with Access Packages. Govern your MCP catalog as a software supply chain — unapproved tools don't run, and approved servers lock behind Azure API Management.
Log every agent tool call, API access, and data lookup into Microsoft Sentinel for continuous anomaly detection. Purview Insider Risk Management auto-assigns risk levels so you can investigate fast or revoke access entirely. DLP and sensitivity labels in Microsoft Purview restrict what agents can reach and auto-inherit to everything they generate, and Data Access Governance maps exactly what each agent can access before a prompt fires.
Jeremy Chapman, Microsoft 365 Director, shares how to put these controls into practice across every managed, self-hosted, and shadow agent in your estate.
► QUICK LINKS:
00:00 - How AI changes Zero Trust
01:20 - Zero Trust principles
02:27 - How to apply Zero Trust principles
03:40 - Conditional Access for Agent Identities
04:59 - Entra Agent ID + Access Packages
06:07 - Runtime Observability
06:58 - DLP, Sensitivity Labels + Data Access Governance
07:47 - MCP catalog
08:36 - AI apps & experiences
09:24 - Wrap up
► Link References
Watch the rest of this series at https://aka.ms/ZTMechanics
For additional resources, check out https://aka.ms/GoZeroTrust
► Unfamiliar with Microsoft Mechanics?
As Microsoft's official video series for IT, you can watch and share valuable content and demos of current and upcoming tech from the people who build it at Microsoft.
• Subscribe to our YouTube: https://www.youtube.com/c/MicrosoftMechanicsSeries
• Talk with other IT Pros, join us on the Microsoft Tech Community: https://techcommunity.microsoft.com/t5/microsoft-mechanics-blog/bg-p/MicrosoftMechanicsBlog
• Watch or listen from anywhere, subscribe to our podcast: https://microsoftmechanics.libsyn.com/podcast
► Keep getting this insider knowledge, join us on social:
• Follow us on Twitter: https://twitter.com/MSFTMechanics
• Share knowledge on LinkedIn: https://www.linkedin.com/company/microsoft-mechanics/
• Enjoy us on Instagram: https://www.instagram.com/msftmechanics/
• Loosen up with us on TikTok: https://www.tiktok.com/@msftmechanics
Published on:
Learn moreWe can help you with Zero Trust security for AI agents
If you want help implementing, troubleshooting, or improving this product, contact us and we’ll point you in the right direction.
Made for tech enthusiasts and IT professionals. Expanded coverage of your favorite technologies across Microsoft; including Office, Azure, Windows and Data Platforms. We'll even bring you broader topics such as device innovation with Surface, machine learning, and predictive analytics.
Related posts
Introducing Autopilot, your personal and proactive agent
Put an AI teammate to work around the clock with Autopilot in Microsoft Copilot, a persistent, proactive and personal agent that runs in the c...
Build apps by describing what you want | Copilot Code
Turn your app idea into a working, interactive web app with Copilot Code. Describe what you want in plain language, ground it in your work dat...
Manage AI Cost + ROI with Copilot and Agent 365
Take control of AI spend across Microsoft Copilot, Cowork, and agents inside and outside of Microsoft. Track your own Copilot Credit usage tas...
Azure SQL Database Hyperscale - Now for any size workload
Start small and scale up to your most demanding workloads on Azure SQL Database Hyperscale. Run serverless compute that scales with demand and...
New Microsoft Copilot Harness | More power, token efficiency & control for agents in Foundry
Build agents you can trust with your company's data, apps, and resources using Microsoft Foundry. Start with a fully managed prompt agent and ...
New Conditional Access policy plans in Microsoft Entra
Group related Conditional Access policies into policy plans and define users, apps and conditions once for every rule in the plan. Add a rule ...
Fight AI with AI | Multi-Agent Security Defense
Match AI-powered attacks at machine speed with Project Perception. Point red, blue, and green AI agent teams at a known threat actor, confirm ...
Build Automated Agents, Workflows, and Apps - New in Copilot Studio
Take a business process from prompt to production by building agents, workflows, and apps in one place with Copilot Studio. Describe the agent...
Create agents fast with Agent Builder in Copilot
Build a specialist AI agent without writing a line of code. Describe what you want in one prompt, and Copilot Agent Builder generates the name...
New Agent 365 controls for Microsoft admins
Govern every AI agent running across your organization with Agent 365. Track agents from Microsoft, Amazon, Google, and Salesforce in one regi...