Skip to content
Theme
OnDevice Tools

WindowsWindows 11

Microsoft explains enterprise governed AI agents on Windows

Microsoft's Windows Blog has outlined how enterprises can deploy governed AI agents using identity, containment and management controls across Windows, Entra, Intune, Defender, Purview and Agent 365.

Windows laptop showing governed AI agent security controls on screen

Key takeaways

  • Microsoft says identity, containment and manageability are foundations for governed AI agents on Windows.
  • Microsoft Entra Agent ID separates agent activity from user activity.
  • Microsoft Execution Containers enforce local boundaries over files, resources and endpoints.
  • Intune, Defender, Purview and Agent 365 provide policy, protection, data controls and visibility.
In this article
  1. Why agent governance matters
  2. Identity, containment and manageability
  3. How Microsoft products fit together
  4. Who is affected
  5. What to do before deployment

Why agent governance matters

Microsoft has outlined how enterprises can deploy governed AI agents on Windows. The Windows Blog post, dated 17 November 2026, says identity, containment and manageability are foundations for safe agent use.

The post says AI agents can improve productivity in enterprise workflows. It also warns that agents can create new risk paths between agents, people, tools and applications.

Microsoft says existing application controls are still important. However, they were not designed to capture the wider context of what an agent does. That context can involve child processes, which are tasks started by other processes, files, network endpoints and long-running sessions.

The post frames the issue as a choice between broad access and blocked value. Microsoft says governed deployment is meant to avoid that trade-off.

Identity, containment and manageability

Illustration: AI, Microsoft logo, business, security and the labels AI Agents, Microsoft, Microsoft Defender

Microsoft says running agents safely at enterprise scale requires three foundations: identity, containment and manageability. Identity should show which agent is acting. It should separate that activity from the user's own actions.

Microsoft Entra Agent ID provides the identity layer for supported agent scenarios. Microsoft Entra governs identity, authentication and access. Windows adds local isolation boundaries. These help keep agent processes and resource access apart from the user.

Containment restricts what an agent can reach during a task. Microsoft Execution Containers, or MXC, provide boundaries enforced by the Windows platform. At runtime, MXC controls what the agent can read or write. It also controls which resources the agent can call. It governs which requests or results can move to permitted endpoints and MCP servers, which connect agents to tools.

Microsoft says MXC should be a key layer in a defence-in-depth design. For a website build, the agent should reach the project folder only. It should not reach private files or unrelated sensitive data. Microsoft says containment strength should match the agent pattern. A coding assistant may need light process isolation. A fully autonomous agent may need stronger isolation, such as separate Windows sessions.

How Microsoft products fit together

Microsoft Intune can apply MXC containment policy to managed Windows devices. It can also control agent access using device compliance, app protection policies for work data and limits on unauthorised agents. It can also apply policy that enables Microsoft Defender runtime protection.

Microsoft Defender can find supported local agents. It supplies endpoint context, such as users, devices, processes and configured MCP connections. Runtime protection can spot and block malicious instructions and tool interactions in supported workflows. Alerts and investigation features help security teams respond.

Microsoft Purview applies data loss prevention policies to supported local-agent interactions. This includes prompts and tool calls. The aim is to spot and block inappropriate sharing of sensitive information.

Microsoft Agent 365 places supported local agents and their MCP connections in a unified registry. It gives administrators a view of the AI estate and the tools agents use. The post says Dynamic Agent Risk, agent templates and integrated security controls can help teams rank risks. They can also help apply governance at scale, manage approved agents and restrict unauthorised agents.

Who is affected

The guidance is aimed at enterprises that use or plan to use local AI agents on Windows. It is relevant to IT teams, security teams and administrators who manage Windows devices, identities and cloud access.

Agents that touch files, call tools, connect to MCP servers or interact with cloud resources need clear boundaries. Microsoft says the goal is consistent governance across endpoints and cloud services. It says fragmented controls are less desirable.

The post also notes that support depends on the agent and scenario. Organisations should check which capabilities apply to their specific agent patterns.

What to do before deployment

Microsoft advises organisations to begin with an inventory of current agent use. It should cover which agents are running. It should also list the identities and credentials they use, who uses them and the workloads they perform.

It should also record the tools and plug-ins each workload calls. It should list the files and directories it accesses and the network endpoints it contacts. This information helps define containment and access policies.

These are vendor-described capabilities from Microsoft's Windows Blog. Organisations should review Microsoft's official guidance before deployment and map controls to their own risk requirements.

Sources

  1. Enterprise Deployment of Governed Agents on WindowsWindows Blog

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.