<!-- Source: https://www.nitronedge.com/insights/microsoft-security-agent-permissions-mcp-discovery/ -->
# Microsoft adds agent permissions and MCP discovery

Microsoft's November security update adds task-scoped agent permissions, Windows agent containment and MCP discovery in Defender. What to do next.

Published 2026-11-12

Microsoft has published its November 2026 security update, and most of it is about one problem: how to keep AI agents in check once they are running inside a company. The roundup, written by Alym Rayani of Microsoft Security and posted on 5 November 2026, covers task-scoped permissions for agents, containment for agents on Windows, and general availability of local AI agent and MCP server discovery in Microsoft Defender.

## What changed

**Dynamic Permissions.** Each agent gets its own permissions, limited to the task it is doing and to how long that task lasts. A feature called Dynamic Agent Risk checks whether that authority still fits how the agent is behaving. If it does not, Dynamic Conditional Access can narrow the permissions, block access to a resource, or ask a person to approve higher-impact actions. Microsoft did not state an availability status for this in the post.

**Microsoft Execution Containers (MXC).** First shown at Build in June, MXC is a containment layer enforced by Windows itself. It applies least-privilege rules to what an agent can do with files, the network, other processes and the user's session, with stronger isolation for riskier cases. The post lists it as generally available. Policies can be tested and tuned before they are enforced widely. Integration with Intune, Entra and Agent 365 is described as coming, not yet here.

**Local AI agent and MCP discovery in Defender.** Declared generally available on 29 October and included in this roundup, discovery works on Windows and macOS. Defender finds more than 50 supported AI agents on endpoints by watching their activity, and links each agent to the device, the user account and any Model Context Protocol servers it is set up to use. Runtime protection for these agents, which can audit or block prompt injection on Windows, remains in preview. Both need Defender for Endpoint P2 or higher.

The update also includes smaller changes. Intune Endpoint Privilege Management lets standard users change network adapter and DNS settings under admin-set rules. Defender Experts MDR has a new overview of active and completed threat hunts. The improved Defender for SQL Servers agent is now generally available in Azure Government without needing Azure Monitor Agent. A Secure Now view in Security Exposure Management links current threat information to clear actions, such as removing excess permissions.

## Why it matters

Agents are spreading faster than most security teams can count them. Developers install coding agents, analysts run desktop assistants, and many of these tools connect to MCP servers that reach email, files, code and business systems. Each of those connections is a path for data to leave or for a prompt injection to cause harm.

This update addresses three separate gaps. Discovery tells you which agents exist and what they connect to. Containment on the device limits what a compromised or confused agent can reach. Task-scoped permissions in identity reduce how much damage any one agent can do, and bring a person back into the loop for actions that matter.

The model is the same one security teams already use for people: know who is acting, give them the least access needed, and watch for behaviour that does not fit. The difference is that agents act faster and in larger numbers, so the controls must be automatic.

Two cautions. First, several pieces are not yet joined up. MXC policies cannot yet be managed through Intune or Entra, so early users will be handling them separately for now. Second, Defender discovery covers supported agents only. Home-grown agents and less common tools may not appear, so discovery should be one input to an agent inventory, not the whole of it.

## What this means for organisations in the Middle East, India, Africa and North America

Many enterprises in the UAE, Saudi Arabia, India, South Africa and Canada already run Microsoft 365, Entra and Defender. For them, these features arrive inside tools they own and staff they have trained, which lowers the cost of starting.

Regulators are beginning to ask about AI systems in the same breath as other critical systems. In the Gulf, the UAE and Saudi data protection laws expect organisations to know which processes touch personal data, and national cybersecurity controls expect asset inventories to be complete. An agent connected to a mailbox through an MCP server is an asset that touches personal data. In India, the DPDP Act puts the duty on the organisation to prevent misuse of personal data, whoever or whatever does the processing. In the USA and Canada, teams using the NIST AI RMF or OSFI E-23 will find agent discovery a practical way to fill in their AI inventory.

Organisations in sectors such as banking, energy and government should also check whether the agent discovery and permission data Defender collects is stored in a region that meets their residency rules.

## What to do next

- Check you hold Defender for Endpoint P2, then turn on local AI agent and MCP discovery for a pilot group and compare the results with what you believe is in use.
- If you try runtime protection, start in audit mode for a week or two before blocking anything.
- Use the findings to start, or update, an inventory of agents with an owner and a business purpose for each.
- Test MXC policies in audit mode on a small set of Windows 11 devices used by developers or analysts who run agents.
- Agree which agent actions need human approval, such as sending external email or changing financial records.

NitronEdge works with Microsoft technologies for security and AI governance. See our [Microsoft Azure partner page](/solutions/agentic-ai-governance/microsoft-azure/) or our [AI security services](/services/ai-security/) for more.

## Sources
- [What's new in Microsoft Security: November 2026](https://www.microsoft.com/en-us/security/blog/2026/11/05/whats-new-in-microsoft-security-november-2026/) (2026-11-05)
- [Now generally available: Discover local AI agents with Microsoft Defender](https://techcommunity.microsoft.com/blog/microsoftthreatprotectionblog/now-generally-available-discover-local-ai-agents-with-microsoft-defender/4568216) (2026-10-29)
