WED, NOVEMBER 18, 2026
Independent · In‑Depth · Practitioner‑Tested
✎ AI Agents

Microsoft Will Issue Identities to Agents It Does Not Own - Starting with Claude Code

Entra Agent ID for discovered local agents starts with GitHub Copilot CLI and Claude Code, separating agent activity from human activity in the audit trail. An MCP Firewall arrives in preview alongside it, controlling which MCP servers, tools and prompts an agent can reach.

By AIToolsRecap November 15, 2026 6 min read 48 views
Home › Articles › AI Agents › Microsoft Is Giving Claude Code an Identity in ...
Microsoft Will Issue Identities to Agents It Does Not Own - Starting with Claude Code
Agent governance at Ignite 2026
● Agent ID for local agents Entra automatically assigns an Agent ID to agents found on managed devices. Microsoft names GitHub Copilot CLI and Claude Code first. Preview.
● MCP Firewall Controls access to MCP servers, tools, resources, prompts, protocols and transports. Preview.
● MCP Server for Enterprise Lets agents reach Entra data through a permissions-based natural-language interface. Generally available.
● Device side Intune discovers agents through Agent 365, marks devices with unapproved agents noncompliant, and Conditional Access blocks them from company data. Preview.

The headline models at Ignite will get the coverage. The more consequential announcement for anyone running agents at work is that Microsoft has started treating them as identities rather than as processes.

Agents get their own ID - including the ones Microsoft did not build

Entra will automatically assign an Agent ID to local agents it discovers, so that what an agent does is separated from what the signed-in human does in the audit trail. The two agents Microsoft names as the starting point are GitHub Copilot CLI and Anthropic's Claude Code.

Naming a competitor's product as a first-class citizen of your identity system is a deliberate choice. It concedes that Claude Code is already running inside enterprises whether or not Microsoft sells it, and that pretending otherwise leaves a hole in the audit log. The same capability extends to Microsoft's own cloud agents, starting with Autopilot agents, also in preview.

The MCP Firewall

Model Context Protocol is how agents reach tools and data, and until now there has been no standard way to control that at the network layer. Microsoft's answer sits inside Global Secure Access: an MCP Firewall, in preview, that governs which MCP servers, tools, resources, prompts, protocols and transports an agent may use.

Prompts being on that list is the detail worth noticing. It means the control surface is not simply which server an agent can dial, but which capabilities on that server it may invoke.

Two related pieces are already generally available: agentic traffic acquisition, which selectively captures agent traffic for visibility and enforcement, and Microsoft MCP Server for Enterprise, which lets agents query Entra data through a permissions-based interface. Local agent network control and an unsanctioned-agents workflow spanning Intune, Defender and Global Secure Access are both in preview.

And on the device

Intune gains local AI agent governance in preview: it discovers agents on managed devices through Agent 365, marks a device carrying an unapproved agent as noncompliant, and lets Conditional Access block that device from company data. Approved agents run inside Microsoft Execution Containers under Intune policy.

Windows adds process container controls in preview, with COM and IPC governance, granular network policy and identity protections. A learning and permissive mode is generally available, so teams can watch what agents actually do before enforcing anything - which is the sensible order, and rarely the one products ship in.

Four overlapping layers is a lot to hold in your head. We have set them side by side with what each one actually stops.

FAQ

Does Entra Agent ID work with non-Microsoft agents?
Yes. Microsoft names GitHub Copilot CLI and Claude Code as the first local agents to receive automatic Agent IDs. The feature is in preview.
What is the MCP Firewall?
A Global Secure Access capability, in preview, that controls which MCP servers, tools, resources, prompts, protocols and transports an agent can reach.
Can I block unapproved agents from company data?
That is the stated design. Intune discovers agents through Agent 365 and marks the device noncompliant, and Conditional Access then blocks it. The capability is in preview.
Is any of this generally available?
Some. Microsoft MCP Server for Enterprise and agentic traffic acquisition are generally available, as is the learning and permissive mode for Windows process containers. Agent ID, MCP Firewall and the Intune governance are in preview.
Tags
AI NewsAI agentsAnthropicClaude Code2026

Spot an inaccuracy?

We verify facts before publishing and correct errors promptly. If something in this article is wrong or outdated, let us know.

Report an error →