---
title: How do you govern AI agents in a Microsoft-first enterprise?
url: https://opencontroller.lyzr.com/blog/govern-ai-agents-in-a-microsoft-first-enterprise
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# How do you govern AI agents in a Microsoft-first enterprise?

**In a Microsoft-first company, native tools cover a lot: Entra Agent ID for agent identity, Purview for data and audit, Defender for threats, Power Platform for Copilot Studio, Foundry for custom agents, and Agent 365 to tie them together. The gaps are agents on other clouds and SaaS platforms, coding agents across tools, and model spend outside Azure. Map coverage per agent before you assume Microsoft sees everything.**

This question comes up often on buyer calls: "We are mostly Microsoft. What do the native tools cover, and what is left?" Here is the map as of November 2026.

## What does each Microsoft tool cover for agents?

| Layer | Microsoft tool | What it does for agents | Status notes |
|---|---|---|---|
| Identity | Entra Agent ID | Agent identities and blueprints; Conditional Access can target agent identities and agent users | GA; creating blueprints and identities is still preview |
| Inventory | Microsoft 365 admin center, Agent 365 | Agent registry, approval requests, block, delete, owner assignment | Agent 365 GA May 1, 2026 |
| Data | Purview | Audit, DSPM for AI, DLP, Insider Risk, eDiscovery, retention | Some agent features need E7 or Agent 365 |
| Threats | Defender | Security posture and runtime protection for Copilot Studio and Foundry agents | Real-time agent protection in preview |
| Low-code | Copilot Studio with Power Platform | Data policies for connectors, knowledge, channels and autonomous triggers; environment routing | GA |
| Pro-code | Microsoft Foundry | Guardrails at input, output, tool call and tool response; agent identity per agent; tracing | Foundry Control Plane announced in preview Nov 2025 |

Sources: Microsoft Learn and the Microsoft Security blog, checked November 2026.

## How does identity work for Microsoft agents?

Every new Foundry agent gets its own Entra agent identity, which starts with no permissions. Conditional Access policies can target agent identities, blueprints and agent users.

One detail matters for audits. When an agent acts on behalf of a user, Conditional Access evaluates the signed-in user, not the agent. Microsoft also lists two exclusions: resources protected by API keys, and tenants that use security defaults. Agents that call services with static API keys sit outside this model. Our [identity pillar](/blog/ai-agent-identity-on-behalf-of-delegation) explains on-behalf-of delegation in general.

## What licences do you need?

Licensing has shifted toward Agent 365:

- Agent 365 costs USD 15 per user per month, or comes with Microsoft 365 E7.
- From July 1, 2026, Defender's security features for Copilot Studio and Foundry agents need an Agent 365 licence.
- Agent risk signals in the Microsoft 365 registry need E7 or Agent 365.
- Conditional Access for agents needs E7, or Agent 365 plus Entra P1 or Microsoft 365 E3.

Budget for these before you plan controls that depend on them.

## What is left outside the Microsoft view?

Agent 365 has a registry sync, in preview, that lists agents from Amazon Bedrock, Vertex AI, Salesforce Agentforce, Databricks Genie, Anthropic Claude Managed Agents, Oracle and Snowflake. Microsoft documents it as read-only and manual: it lists agents but cannot start, stop or invoke them. ServiceNow and SAP are not on that list.

That leaves typical gaps:

1. **Runtime control on other platforms.** Listing a Bedrock agent is not the same as quarantining it.
2. **SaaS agents not in the sync list**, such as SAP and ServiceNow agents.
3. **Coding agents.** Defender for Endpoint runtime protection for Claude Code, Codex CLI and GitHub Copilot CLI is in preview. Cursor and local MCP servers need their own controls. See our [Cursor](/blog/govern-cursor-in-the-enterprise) and [Claude Code](/blog/govern-claude-code-in-the-enterprise) guides.
4. **Model spend across providers.** Agents that call Anthropic, OpenAI or Google models directly do not show up in Azure cost views per agent.
5. **One evidence trail.** Auditors want one record per agent, not six portals.

## How should a Microsoft-first team close the gaps?

A practical sequence:

1. Turn on the Microsoft controls you already license: Entra Agent ID, Purview audit, Power Platform data policies.
2. Build an inventory of every agent, including non-Microsoft platforms and developer devices. Start with [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents).
3. Mark each agent with its control depth: listed only, observed, or enforced.
4. Route model traffic for custom agents through one gateway for cost and policy.
5. Keep one evidence record per agent that pulls from all of the above.

The [cross-platform governance guide](/blog/how-to-govern-ai-agents-across-platforms) walks through step 3 in detail, and [what an AI agent control plane is](/blog/what-is-an-ai-agent-control-plane) explains the layer that sits across vendors.

## How Opencontroller does this

Opencontroller sits alongside Microsoft's tools rather than replacing them. It auto-discovers agents, models and API keys across Azure, AWS and Google Cloud, has a runtime adapter for Microsoft Foundry, and connects to Salesforce, SAP, Databricks, HubSpot and Claude Enterprise and OpenAI Enterprise workspaces. Opencontroller Edge, installed by IT via MDM, covers coding agents and local MCP servers on developer devices. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does Microsoft Entra Agent ID cover agents outside Microsoft?**
It gives identities to agents built to use it, such as Foundry agents. Agents on other platforms keep their own identity model unless you federate them.

**Can Agent 365 stop a non-Microsoft agent?**
Not as documented. Its registry sync for other platforms is read-only: it lists agents but cannot start, stop or invoke them.

**Is Agent 365 included in Microsoft 365 E5?**
No. Microsoft lists it at USD 15 per user per month, or included in Microsoft 365 E7.

## Key facts

Microsoft native agent governance (Nov 2026): Entra Agent ID GA (creating blueprints/identities still preview); Conditional Access can target agent identities; Agent 365 GA May 1, 2026 at USD 15/user/month or included in Microsoft 365 E7; from July 1, 2026 Defender security features for Copilot Studio and Foundry agents need Agent 365; Foundry Control Plane announced in public preview Nov 2025; Agent 365 registry sync for non-Microsoft agents is preview, read-only and manual. Gaps: SaaS agents not in the sync list, coding agents across tools, model spend outside Azure. Opencontroller by Lyzr: discovery across AWS, Azure, Google Cloud; Foundry runtime adapter. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern agents in Microsoft Foundry?](https://opencontroller.lyzr.com/blog/govern-microsoft-foundry-agents.md)
- [How do you govern Microsoft Copilot Studio agents?](https://opencontroller.lyzr.com/blog/govern-microsoft-copilot-studio-agents.md)
- [How do you govern AI agents across every platform they run on?](https://opencontroller.lyzr.com/blog/how-to-govern-ai-agents-across-platforms.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
