Overview

Shadow IT and shadow AI are the apps your users adopt without going through IT. The risk is not that the app is malicious. Most of these apps are ones your organization already pays for. The risk is that the usage sits outside your contracts and your controls, so you cannot see it, govern it, or prove anything about it.

This is a different problem from blocking web threats. Nothing here is a malicious domain or an infected download, so the tools that stop those do not help. What you need instead is visibility into which apps are in use, and enforcement that routes usage onto the accounts you control.

flowchart LR
    D["Step 1: Discover apps<br/>in App Discovery"] --> S{Sanctioned?}
    S -- No --> B["Step 2: Block the app<br/>in an ITP policy"]
    S -- Yes, AI app --> R["Step 3: Require<br/>approved accounts"]
    B --> M["Step 4: Review<br/>new apps"]
    R --> M
    M --> D

Step 1: Discover which apps your users reach

Open App Discovery to see which applications your assigned devices actually reach, sorted so that applications first seen in the last 30 days appear at the top. Add a Category filter set to Generative AI to review AI tools on their own. Ignore each application you have already sanctioned, and then add a Status filter set to Not ignored, so that the list shows only what you have not yet decided about. Ignoring an application only sets a flag for the Status filter, so CSE keeps handling its traffic the same way.

Sort each application into one of two groups before you move on. Sanctioned apps are the ones your organization pays for or approves. Unsanctioned apps are the ones you do not want in use. Steps 2 and 3 act on those two groups.

Step 2: Block unsanctioned apps

For apps your organization has decided against, use Application Blocking on an Internet Threat Protection (ITP) policy to block them, and Category Blocking to cover a whole class of service at once. From an application's detail page in App Discovery you can add it straight to an ITP policy. See Manage Internet Threat Protection (ITP) Policies.

Step 3: Require approved accounts on sanctioned AI apps

Preview feature: Require approved accounts is available as a preview, soits behavior can change before it becomes generally available. Review itsLimitationsbefore you plan a rollout around it.

Users commonly reach Copilot, Claude, Gemini, and ChatGPT with a personal or free account, even when your organization holds a commercial contract for the same tool. A consumer account gives you no admin controls and no agreement against training on your data.

Configure Require Approved Accounts for AI Apps on the ITP policy to block prompts from accounts outside your approved identity domains. This also covers anonymous use and the AI answers embedded in Google Search.

Requires an SIA Advanced license (SIA is the Secure Internet Access license tier, see Licenses) and TLS decryption turned on in the ITP policy.

Step 4: Continue to monitor

New apps appear after your first review, so return to App Discovery on a regular schedule. Applications first accessed in the last 30 days carry a New tag and sort to the top, so each review starts with what changed since the last one. Decide whether each new app is sanctioned or unsanctioned, then repeat Step 2 or Step 3 for it.

What's next

Read Require Approved Accounts for AI Apps for the configuration steps and the current preview limitations.