Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Microsoft Mechanics Podcast

Agent 365 | Identity & Access Controls in Entra

Jun 9, 20269 minVideo
ShareXLinkedInFacebookWhatsAppEmail

Take control of every AI agent, managed or not, running in your environment using Agent 365 and Microsoft Entra. Surface agents across AWS Bedrock, Google Vertex, Databricks, and Salesforce in one registry, assign Entra Agent IDs via CLI or SDK, and enforce least-privilege access through Conditional Access policies and Agent Blueprints, all without rebuilding your existing identity infrastructure. Lock down agent activity with sign-in logs that capture every authentication attempt, policy hit, and failure. Govern agents as first-class identities alongside your users, apps, and devices, and draw a hard line between managed and unmanaged AI in your organization. Vince Smith, Microsoft Entra Principal Product Manager, shares how to establish full visibility, access control, and lifecycle governance for AI agents using Microsoft Entra and Agent 365. ► QUICK LINKS: 00:00 - Visibility and control with Agent 365 01:39 - Multi-platform registry sync 02:29 - Assign Agent ID 04:14 - Agent Blueprints 05:24 - Conditional Access for agents 06:24 - Sign-in logs audit trail 07:03 - Unblock the agent 07:54 - Wrap up ► Link References Check out https://aka.ms/EntraforAgents ► 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

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show.

0:03Ensuring agents don't operate unchecked in your environment means making sure their access to data and apps, along with other agents and resources, is appropriately scoped, and no more than necessary, so they're never overprivileged. Agent 365 gives IT and security teams a unified control plane for agent activity. They can work together using the tools they already work in every day to see which agents are in use, understand the risk, and act before issues escalate. Today, in part 3 of this Agent 365 series,

0:34we look at how, as an identity admin using Microsoft Entra, you can establish the critical foundation to prevent agents from being both overprivileged or running without the right level of visibility. This all starts with Agent 365, which gives you visibility into your agents, control over what they can access, and governance across the entire lifecycle. As a developer, you can use the Agent 365 CLI and SDK to create and integrate an Agent ID into your agent. This ensures your agents can be managed

1:04across their lifecycle, with continuous risk evaluation and least privileged conditional access controls in Microsoft Entra, ensuring that agents can only connect to and act on the resources they truly need, for as long as they need, no more, no less. Just like with your users and apps today, with Agent ID, agents become explicitly manageable and governable, including conditional access policies and access packages. And by assigning an Entra identity to your agents, you establish a clear boundary between managed and unmanaged AI, which makes it much easier

1:36to detect which agents need to be brought under control. Let's start in the Microsoft 365 Admin Center to see this in action. As an IT admin, you can access Agent 365 controls to view your agents, find unmanaged AI in use, and configure tools and settings. I'm in the Agent 365 overview page, where I can find top-level insights and metrics, as well as common actions. In fact, the Registry sync controls let me configure these options, so let's take a look. And you can see that we've already configured sync

2:06for AWS Bedrock and Google Vertex, which gives us visibility into agents running on those platforms. Just to show you what's behind this and your options, I'll click Connect a platform. And under External platforms, we can see Bedrock and Vertex here, along with Databricks Genie and Salesforce AgentForce. The respective agents will peer in the agent registry once I sync from these platforms. That said, while we were able to see the agents from those platforms running in our environment, without respective Agent IDs, we won't have visibility

2:37into exactly what those agents are accessing. In fact, here we can see that all of our Amazon Bedrock agents are currently unmanaged. And when I click in, we can see a few details about the agent creation date and publisher, but not much else. So we know these agents exist, but again, without an Agent ID, we don't know exactly what these agents are accessing to be able to apply the right controls for their needs. Practically speaking, our first step here is to work with the unmanaged agent creators to get their agents an Entra ID. This is where, as a developer creating and updating agents,

3:09we can use the agent 365 CLI and SDK to create an assign and Agent ID, which also includes an Agent Blueprint. I've used a simple prompt here to generate the code and enter configuration needed. The cool thing about Agent Blueprints is that permissions granted on the blueprint can be inherited by all agent IDs using that blueprint, and I'll explain more about this in a moment. And because this is standard OAuth, it is interoperable across agent platforms. In fact, the agent we're updating is the one I showed earlier that's running on AWS Bedrock.

3:40And once the agent is republished to Agent 365, it's ready for management. In fact, going back to the agent registry in Agent 365, viewing the details of the agent, we can move over to the right and expand this field for identity information. We can see it has an Agent ID along with a Blueprint ID, which wasn't there when we looked at the unmanaged agent card before. And scrolling back up and opening the Security tab, I can see that the agent already has default protections applied from Microsoft Purview and Entra, including sensitive data protections and compliance evaluation as part of the Agent 365 platform.

4:14Because the agent is an object in our directory, provided we are entitled with the right management permissions, we can grant the appropriate level of access and get visibility into its details and activities. I'll start by first looking at what resources the developer has configured for the agent to access. In the agent identities page, we can see our agent. Importantly, this agent now has an agent sponsor assigned. This is the person responsible for the agent, and can be the agent builder themselves or someone they designate. And viewing its access and granted permissions, we can see a list of the requested permissions

4:46the agent has configured. Notice that the grant source is inherited from the parent, which is our Agent Blueprint. A blueprint is the template for a class of agents. It holds the credentials that associated agents use for authentication. The credentials can be managed in Entra, or federated outside of Entra, or even system managed, such as with an Azure Managed Identity. It also provides a way to define identity and permissions for agents at scale. Here we see in this case, the agent's permissions aren't set directly on the agent itself. They come from its parent blueprint.

5:16The first permission for Agent365.Observability is a default to capture agent telemetry. And under that, you'll see Microsoft Graph permissions to read groups, users, and mail. Under Developer settings in the Manifest, we can view and optionally edit the details as JSON that we saw earlier in the code. As an Entra admin, I can control who is able to consent to the agent's permissions, which gives me both security and scale, but I also want defense in depth. In our organization, as a blanket control, we have conditional access policies

5:48to block access to all unverified agents. Under conditional access, when I click into the AWS Bedrock policy, you'll notice that this policy targets all agent identities and blocks their access to Entra-managed resources by default. To give our now verified agent access, we'll need to exclude it from this conditional access block policy. Before we do that, I'll show you this policy in action from the agent testing experience in Amazon Bedrock. I'll have the agent perform a simple look up for a user account in our tenant. And you'll see that it's blocked

6:18from retrieving the user information. In fact, if I move over to this agent's logs, I can see all the invalid grant errors. And expanding on the first one and then scrolling down to the error description, it gives more specific detail about the conditional access block and what happened. Now, if I move back to Microsoft Entra and look at the sign-in logs, we can see all of the sign-in activities for this agent. I'll expand the same request we just saw in the Bedrock portal and move into this specific instance. And you'll see the failure reason is exactly what the agent developers saw in Bedrock.

6:51In fact, digging into this Conditional Access tab, we can see the corresponding policies listed that apply to this agent, along with AWS Bedrock policy on top and the corresponding failure result. Now let's look at how we can unblock this agent. As a Microsoft Entra administrator, once we verify that this agent's identity is established and its permissions are appropriate, we can exclude it from our conditional access block policy to allow it access. Back in our conditional access policy settings, under Assignments, I can set up an exclude condition. Then I search for my agent by its name.

7:23Here I'm typing, "widget." There it is. Now I'll select it and confirm, then I just need to save my policy. That's it. With our exclusion setup, the agent will be able to perform its operation with least privileged access and Entra. So let's test it out. Back in the Amazon Bedrock portal, I'll run the same prompt that failed last time, our account look up. And as you can see, this time it works, and it was able to access and display the account information for this user from our tenant. There we have it. That's how easy it is to integrate Entra Agent ID into your agents and bring them under management.

7:54Agent 365 with identity and access controls in Microsoft Entra provides IT and security teams the management and visibility to assess and respond to agent risks. This lets you efficiently integrate agents into your existing systems and infrastructure as first-class identities, alongside existing users, applications, and devices. To learn more, check out aka.ms/EntraforAgents. Keep checking back to Microsoft Mechanics for the latest tech updates, and thanks for watching.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Introducing Autopilot, your personal and proactive agent

    Nov 20, 202610 minVideo

    Put an AI teammate to work around the clock with Autopilot in Microsoft Copilot, a persistent, proactive and personal agent that runs in the cloud and keeps working after you power off your laptop. Describe your goals...

    Credit the people on this episode

  2. Build apps by describing what you want | Copilot Code

    Nov 19, 20268 minVideo

    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 data with Work IQ, build what-if modeling from your own business logic, and...

    Credit the people on this episode

  3. Manage AI Cost + ROI with Copilot and Agent 365

    Nov 19, 202610 minVideo

    Take control of AI spend across Microsoft Copilot, Cowork, and agents inside and outside of Microsoft. Track your own Copilot Credit usage task by task, pin down your top cost drivers by service and model provider in...

    Credit the people on this episode

  4. Azure SQL Database Hyperscale - Now for any size workload

    Nov 18, 202614 minVideo

    Start small and scale up to your most demanding workloads on Azure SQL Database Hyperscale. Run serverless compute that scales with demand and auto-pauses, so you stop paying for idle compute. Pool databases across...

    Credit the people on this episode

  5. New Microsoft Copilot Harness | More power, token efficiency & control for agents in Foundry

    Nov 10, 202615 minVideo

    Build agents you can trust with your company's data, apps, and resources using Microsoft Foundry. Start with a fully managed prompt agent and move it onto the Microsoft Copilot harness so it plans, loops, calls...

    Credit the people on this episode