Blog · October 31, 2026

Shadow AI: What It Is, Real Examples, and How Small Offices Find It

Shadow AI is AI that people at work use without the business knowing about it or agreeing to it. IBM defines it as "the unsanctioned use of any artificial intelligence (AI) tool or application by employees or end users without the formal approval or oversight of the information technology (IT) department."

In a large company, "IT" is a department. In a small office, it may be the owner, an office manager or an outside provider. The idea is the same: an AI tool is doing work with company information, and nobody responsible for that information has said yes.

This guide explains what shadow AI is, how it differs from shadow IT, why it happens, the risks in plain terms, and practical ways a small business can find it and set rules. It also covers a part most guides skip: AI agents that run on the Windows PCs in your office.

What is shadow AI? Common examples

Shadow AI is usually ordinary, well-meant work. Typical examples:

Pasting a customer email into a public chatbot to draft a reply.
Uploading a spreadsheet to an AI tool to summarize sales figures.
Using an AI note-taker in client calls.
Installing a browser extension that rewrites text or summarizes pages.
Turning on an AI feature inside an app the office already pays for.
Running an AI agent on a work PC that can open files, use the browser or act in signed-in accounts.

IBM’s list of examples also includes cases where "employees might use external machine learning models to analyze and find patterns within company data."

None of these is wrong on its own. Each one becomes shadow AI when the business didn’t know about it and never decided whether it was OK.

Shadow AI vs shadow IT

Shadow IT is the older, broader idea. IBM describes it as "the deployment of any software, hardware or information technology on an enterprise network without an IT department or CIO’s approval, knowledge or oversight." It then draws the line: "While shadow IT focuses on any unauthorized application or service, shadow AI zeros in on AI-specific tools, platforms and use cases."

	Shadow IT	Shadow AI
What it covers	Any unapproved software, device or service	Unapproved AI tools, features and agents
Typical example	A personal file-sharing account for work files	Pasting client data into a chatbot
Main question	Where is our data stored?	Where does our data go, and what does the AI do with it or on our behalf?
What’s new	—	AI tools can generate content and, with agents, take actions

The key difference for a small business is the last row. A file-sharing app stores what you put in it. An AI tool may process it, and an AI agent may act on it: send, edit, click or buy.

Why shadow AI happens

Shadow AI is mostly a sign that people want to work faster. IBM puts it this way: "Employees often turn to these tools to enhance productivity and expedite processes," and they "might turn to unsanctioned AI technology when they find existing solutions insufficient or believe that the approved options are too slow."

In small offices, common reasons include:

No rule exists yet, so people assume it’s fine.
The tools are easy to start. Many need only a browser and an email address.
AI arrives inside other software, often through a product update.
Nobody owns the question. Without an IT team, deciding what is approved falls between the cracks.

This is why the fix is rarely about blame. It is mostly about visibility and clear rules. How to talk with staff who already use AI tools without permission is its own topic, and we’ll cover it in a separate guide.

Shadow AI risks, in plain terms

We won’t use scare numbers here. These are the risks named by primary sources, explained simply.

Sensitive information leaving the business. OWASP lists Sensitive Information Disclosure in its Top 10 for LLM applications and notes that "sensitive information can affect both the LLM and its application context." In practice: client details, financials or passwords pasted into a tool you don’t control.
Agents with too much access. OWASP’s Excessive Agency entry describes "the vulnerability that enables damaging actions to be performed in response to unexpected, ambiguous or manipulated outputs from an LLM." An unapproved agent signed in to work accounts is the clearest example.
Decisions nobody can explain. If AI output feeds into quotes, contracts or customer replies and nobody knows which tool was used, mistakes are hard to trace.
Obligations you can’t check. Clients, insurers or regulators may ask how you handle data. You can’t answer for tools you don’t know about.
No plan when something goes wrong. If an unknown tool misbehaves, there’s no owner, no setting to change and no record.

For a structured way to think about AI risk, the NIST AI Risk Management Framework "is intended for voluntary use" and helps organizations build trustworthiness into how they use AI. Our AI governance for small business guide turns that into small-office steps.

How to find shadow AI in a small office

You don’t need special software to start. A simple, friendly audit goes a long way.

1. Ask first

Send a short, no-blame question: "Which AI tools, extensions or features do you use for work, and for what?" People usually answer honestly when they know the goal is to set sensible rules, not to punish.

2. Check browsers

Look at installed extensions in each browser on each PC. Microsoft and Google both document where to see and manage extensions in Microsoft Edge and Google Chrome.

3. Check each Windows PC

Look at installed apps, startup apps and what’s running in Task Manager. Our step-by-step guide on how to tell if an AI agent is running on your Windows PC covers each place to look.

4. Check AI features in software you already use

Many business apps now include AI features. Review the settings of your office suite, email, CRM and help desk tools, and note what is switched on.

5. Check connected accounts

AI tools often connect to email, calendars or cloud storage. Review third-party app connections in each account’s security settings.

6. For larger or managed setups

This step is only for organizations that already have Microsoft Defender for Cloud Apps and an admin to run it. You don’t need it to start. For those organizations, Microsoft’s documentation describes a Cloud discovery page that "provides a dashboard designed to give you more insight into how cloud apps are being used in your organization."

Where AI agents on a Windows PC fit

Most shadow AI guides focus on websites and cloud apps. That leaves out AI agents that run on the PC itself, such as desktop assistants, coding agents, browser agents and AI features built into Windows.

These matter for two reasons. First, an agent on a PC can act, not just answer: it may open files, use the browser where staff are signed in, or run commands. Second, cloud-focused tools may not show them, because some of their work happens on the device.

If you’re new to how agents work, our guide AI agents explained covers the main types and where you meet them. To check a specific PC, use how to tell if an AI agent is running.

Set rules so AI use stops being "shadow"

The goal isn’t zero AI. It’s AI the business knows about and has agreed to. A short written policy does most of the work:

List approved tools and what each may be used for.
Name data that never goes into AI tools, such as client personal data, passwords and financial records.
Set a simple way to request a new tool, with a quick yes or no.
Name one owner who keeps the list current.
Add rules for agents, covering what they may access and when a person must approve an action.
Review a few times a year, because tools and features change quickly.

Start with our AI acceptable use policy template for small business, then add agent-specific rules from our AI agent policy for small business guide.

How PERCARIS helps

Rules work best when you can see what’s happening on each PC. PERCARIS surfaces recognized AI-agent activity on your Windows computer and gives you a Stop control for agents you did not authorize. PERCARIS recognizes known AI tools. It won’t flag every AI agent.

You decide whether to press Stop. PERCARIS works alongside your existing security tools, and each Windows PC is managed on its own. You can read more on how it works.

Plans cover a single PC, a few PCs or a small business. For 26 or more PCs, contact support@percaris.com. See plans.

FAQ

What is shadow AI?

Shadow AI is the use of AI tools, features or agents at work without the business's knowledge or approval. IBM defines it as unsanctioned AI use "without the formal approval or oversight" of IT.

What is the difference between shadow AI and shadow IT?

Shadow IT covers any unapproved software, hardware or service. Shadow AI is the AI-specific part of it, where the tools can generate content and, with agents, take actions.

What are the main shadow AI risks?

Sensitive information leaving the business, agents with too much access, AI-driven decisions nobody can trace, and obligations to clients or regulators you can't check.

How can a small business find shadow AI?

Ask staff first, then check browser extensions, installed and running apps on each PC, AI features in existing software, and connected accounts.

Should we ban AI tools to stop shadow AI?

Most small businesses do better with a short list of approved tools and clear rules about data. Bans often push AI use further out of sight.

Sources

Checked against each source on October 31, 2026. AI tools and definitions change often.

— The PERCARIS Team