Shadow AI agents are AI tools your employees install and run on company devices without IT approval. This guide to Shadow AI Agents: How to Find and Block Them with Intune shows IT administrators how to use Microsoft Intune, Agent 365, and Microsoft Defender for Endpoint to see which local AI agents are running, who runs them, and how to stop the ones you don’t allow.
Local AI agents can read local files, run commands, and act with the user’s own permissions. That makes them useful, and it also makes them a blind spot. Microsoft now gives endpoint teams built-in Shadow AI detection and blocking, so you no longer have to rely on custom scripts or hand-written discovery rules.
What Are Shadow AI Agents, and Why Are They a Risk?
Shadow AI means consumer-facing AI applications and standalone agents running in your organization without IT visibility or approval, as Microsoft defines it. These unmanaged AI agents are different from a browser chatbot. They run on the endpoint, so network and identity controls alone often miss them.
A local agent can start from the command line, a desktop app, an IDE extension, Node.js, or Windows Subsystem for Linux. Once running, it can inherit access to local files, code repositories, tokens, and enterprise services from the signed-in user.
Microsoft lists four main risks of unmanaged agents:
Data leakage: company data sent to outside AI services.
Compliance violations: data handled outside approved processes.
Security vulnerabilities: outdated or untrusted agents with broad local access.
Lack of auditability and governance: no record of what the agent did.
For regulated organizations in the United States, from healthcare to finance, that last point matters most. You can’t prove control over data flows you can’t see.
Common local AI agents you may find
| Type | Examples |
| Command line agents | OpenClaw, Claude Code CLI, Codex CLI, GitHub Copilot CLI, OpenCode |
| Desktop apps | ChatGPT Desktop, Claude Desktop, Codex Desktop, Ollama Desktop, Poe Desktop |
| Agentic IDEs | Cursor |
| Claw-based agents | Claw/ZeroClaw |
On a phone, scroll the table sideways to see both columns.
Microsoft’s Shadow AI Toolkit: Agent 365, Intune, and Defender
Microsoft now covers shadow AI with three connected tools. Each answers a different question.
| Tool | Where | What it does |
| Shadow AI page (Agent 365) | Microsoft 365 admin center | Detects known shadow agents and blocks OpenClaw with one click |
| Local AI agent inventory | Microsoft Intune | Collects agent details through the Properties catalog, Device Inventory, and Device Query |
| Local AI agent discovery | Microsoft Defender for Endpoint | Finds 20+ supported agents and their MCP configurations, and powers Shadow AI page detection |
All three are in public preview. The Shadow AI page is part of the Frontier preview program, so features and supported agents may change. At Ignite 2026, Microsoft also previewed an Intune software compliance policy that can mark devices running unsanctioned agents as noncompliant and use Conditional Access to block their access to corporate resources, per the Intune Ignite announcement.
What you need
License: Microsoft’s current docs list a Microsoft 365 E5 license to view Shadow AI agents. Earlier versions listed a Microsoft 365 E3 license, so confirm before you buy.
Devices: Windows 10 and later, enrolled in Intune. The Properties catalog needs corporate-owned devices that are Microsoft Entra joined or hybrid joined.
Defender: Microsoft Defender for Endpoint enabled on devices for shadow agent detection.
Frontier: Opt in to the Frontier preview program in the Microsoft 365 admin center.
Optional: Global Secure Access, to see network data about each agent.
Role: At least one of Security Administrator, AI Administrator, Global Reader, Security Reader, Security Operator, Reports Reader, User Experience Success Manager, or Intune Administrator.
How to Find Shadow AI Agents
Start with discovery, not blocking. You need evidence of what’s running before you decide what to stop.
Option 1: The Shadow AI page in the Microsoft 365 admin center
Sign in to the Microsoft 365 admin center.
Select Agents > Shadow AI to open the Shadow AI page.
Select an agent, such as OpenClaw, to open its details.
In earlier preview builds, you turn on detection under Security policies > Continuously detect managed devices.
The details pane shows when the agent was first accessed, its most recent activity, and when it was last scanned. It also shows how many devices and users run it, and a security policies tile that tells you if any Microsoft Intune security policies, such as a block, are already applied.
Open the Detected devices tab to see each device name, model, operating system, and the last time Microsoft Defender detected the agent on that device.
If you turn on Global Secure Access, the Total traffic tile adds network detail: unique endpoints the agent reached, network requests made, data sent, and data received.
Option 2: Local AI agent inventory in Intune
Intune gives you deeper detail on every agent, as Jannik Reinhard shows.
In the Intune admin center, go to Devices > Configuration > Create > New Policy.
Choose Windows 10 and later and the Properties catalog profile type.
Under Add properties, search for “AI agent” and add the Local AI Agent (Preview) category.
Assign the profile to a small pilot group first. Data can take up to 24 hours to appear.
To check one device, open Monitor > Device Inventory. For the whole fleet, use Device Query and pick the Local AI Agent entity.
What to look at for each agent
| Field | Why it matters |
| Agent name and version | Spot unapproved or outdated agents |
| Host process | Shows if it runs through Node.js, WSL, or another runtime |
| Install location | Separates managed installs from user-local copies |
| Install scope | Per-user or per-device install |
| User identifier | Who runs it, and for what business reason |
| Execution context | How it launches and what access it inherits |
Microsoft recommends collecting the host process, because OpenClaw can run under different processes such as node.exe or wsl.exe, according to Microsoft Learn.
Option 3: Microsoft Defender for Endpoint
Defender goes furthest on discovery and investigation. Under Assets > AI agents, it lists supported desktop apps, CLI agents, agentic IDEs, and their MCP configurations. That helps you see what each agent can reach, not just that it’s installed.
Classify what you find
Sort each agent into one of four groups: approved, approved with conditions, unknown, or prohibited. This works better than a single allow-or-block list, because each tool has a different risk and business value.
Do you know which AI agents are running on your managed devices today?
We can set up discovery in Intune, Defender and Agent 365 for a pilot group and help you classify what turns up.
How to Block Shadow AI Agents with Intune
Today, blocking a Shadow AI agent from the Shadow AI page works for OpenClaw only. The other agents on the list are detection only, so you’ll govern them through policy, Defender, and user guidance for now.
Block OpenClaw from the Shadow AI page
Open the OpenClaw details pane on the Shadow AI page.
Select Security policies.
Select Block AI agents from OpenClaw (in newer builds, Block > Apply Policies).
This creates an Intune policy named A365 – Block OpenClaw that rolls out to all managed Windows devices. It blocks common execution methods for the agent. Depending on your Intune setup, the policy takes anywhere from 15 minutes up to 8 hours to apply. You can find and edit the policy in Intune to add more control.
Or use the Intune security baseline
In Intune Endpoint security, the Local AI Agent Baseline – OpenClaw does the same job with more visibility into each setting. Its settings target Windows Subsystem for Linux and add outbound firewall rules for Node.js executables.
Test it before broad rollout. Node.js and WSL are not unique to OpenClaw, so a broad block can break developer tools and legitimate apps. Microsoft also notes the baseline may not block every execution path, so treat it as one layer, not a complete fix.
Block access, not just the app
The new Intune software compliance policy, in public preview, takes a different approach. Instead of stopping the app, it marks the device as noncompliant when an unsanctioned agent is present. Conditional Access then blocks that device from company data until the agent is removed. This helps reduce data exposure and maintain compliance even for agents the Shadow AI page can’t block yet.
A 6-Step Rollout Plan for Endpoint Teams
Treat every shadow AI control as a controlled pilot while it’s in preview.
Check prerequisites. Confirm licensing, Intune enrollment, Defender onboarding, Frontier opt-in, and admin roles.
Turn on discovery. Deploy the Local AI agent inventory profile to a pilot group: developer workstations, IT admin devices, and one clean reference device with no AI tools.
Investigate the results. Review host process, install location, user, and execution context for every agent, then classify each one.
Build an exception list. Document approved Node.js and WSL workloads before you block anything.
Block in stages. Apply the OpenClaw block or baseline to the pilot, watch for broken apps and network impact, then expand.
Review weekly. Check the Shadow AI page for newly detected agents, since Microsoft keeps adding support.
Once it’s running, this process can significantly reduce the need for manual effort compared with tracking AI tools through spreadsheets and user surveys.
Frequently Asked Questions
Which agents can the Shadow AI page block?
Only OpenClaw today. ChatGPT Desktop, Claude Desktop, Ollama Desktop, Poe Desktop, Claw/ZeroClaw, and OpenCode are detection only during the public preview.
Does Shadow AI blocking work on personal devices or Macs?
No. Shadow AI blocking applies only to managed Windows devices enrolled in Intune. Personal devices and anything outside Intune stay invisible to the page.
How long does a block take to apply?
The A365 – Block OpenClaw policy takes from 15 minutes up to 8 hours, depending on your Intune configuration.
Will blocking OpenClaw break other apps?
It can. The baseline restricts WSL and Node.js paths that other developer tools also use. Pilot first and keep an exception list.
Do I need Global Secure Access?
No, but without it the Total traffic tile and some activity fields stay empty.
Get Ahead of Shadow AI Before It Becomes an Incident
Shadow AI isn’t going away. Your people will keep trying new agents because they save time. The goal isn’t to block everything. It’s to know what’s running, approve what’s safe, and control the rest.
Windows Management Experts helps organizations set up Intune, Defender, and Agent 365 for shadow AI discovery and run a safe controlled pilot. Talk to a WME consultant to get an AI agent exposure assessment for your Intune environment.
Know what’s running, approve what’s safe, and control the rest.
We’ll set up Intune, Defender and Agent 365 for shadow AI discovery and run a safe, controlled pilot.


