What is shadow AI, and why it matters for small teams

Someone on your team pastes a client email into a chatbot to help write a reply. Someone else uses a free AI tool to summarise meeting notes. Nobody asked, and nobody said no. That is shadow AI.

What shadow AI means

Shadow AI is any use of AI tools for work that the business has not chosen, approved or even heard about. The name comes from an older idea, “shadow IT”, which describes apps and devices people bring into work on their own because they are quicker than the official route.

It almost always starts with good intentions. AI tools are easy to sign up for, often free, and genuinely helpful. People use them because they want to do good work faster. So this is a conversation about visibility, and nobody needs to be in trouble.

Why it matters for small teams

Small teams move fast and run on trust. That is a strength. It also means there is often no one whose job is to notice which tools are handling your information.

Where your information goes

When you paste text into an AI tool, it is sent to that company’s servers. Depending on the tool, the plan and your settings, it may be stored, reviewed by people, or used to improve the company’s AI models. Free and personal plans often have different rules from business plans. If that text includes client details, health or financial information, or anything covered by a confidentiality agreement, you may be sharing more than you meant to.

Who owns the account

A tool someone signed up for with their personal email belongs to that person. If they leave, the chat history, files and settings leave with them, and the business has no way to review or delete what was shared.

What it is connected to

Some AI tools ask to connect to your email, calendar or shared files. Browser extensions (small add-ons to your web browser) can often read the pages you visit. Each connection is a door into your business, and it is easy to click “Allow” without reading what you are agreeing to.

What goes out with your name on it

AI tools can sound confident and still be wrong. A summary that misses a detail, or a reply that invents a policy, can reach a customer if no one checks it first.

What to do about it, without banning everything

Banning AI rarely works. People tend to keep using the tools, just more quietly. A lighter approach works better.

Ask, kindly. Send a short, no-blame question: which AI tools do you use for work, and what for? You will learn a lot in a day.
Choose a few approved tools. Where you can, use business or team plans. They usually include admin controls, and many let you keep your data out of model training. Read the provider’s privacy and data terms, and check the settings yourself.
Write a one-page policy. Keep it simple: which tools are approved, what they are fine to use for, and what never goes in. A good starting list for “never” is passwords, payment details, personal information about clients or staff, and anything under a confidentiality agreement.
Name an owner. One person keeps the list of approved tools, answers questions and says yes or no to new ones. In a small team, this might take an hour a month.
Review connected apps. Check which apps have access to your email and files, and remove the ones you no longer use.
Check in regularly. New tools appear all the time. A quick review every few months keeps your list current.

One more thing helps a lot: make the approved route the easy route. If the approved tool is already set up and ready to go, people will reach for it first.

Don't just ask. See it for yourself.

Asking works well in a small, close team. But people forget, new tools appear every week, and businesses that handle health, legal or financial information often need proof, not just trust. The good news: the cloud platforms you may already pay for include tools that find AI use automatically and enforce your rules. Here's how that looks with Google.

1. Find where AI is being used

Google Workspace Admin console. App access control lists every outside app connected to your team's Google accounts, including AI tools that asked to read Gmail or Drive. You can block the ones you haven't approved.
Chrome Enterprise. With managed Chrome browsers, you can see which extensions are installed, including AI add-ons, across your team.
Security Command Center. In Google Cloud, AI Protection builds an inventory of the AI models and agents running in your projects and flags risky settings.
Cloud Audit Logs. A record of who used which AI services in your cloud projects, and when.

2. Turn your policy into settings

Workspace AI controls. Switch Gemini and other AI features on or off for specific teams.
Chrome allowlists. Allow only approved AI websites and extensions, and block the rest.
Organization Policy. In Google Cloud, limit which services, including AI services, your projects can switch on.

3. Protect sensitive information automatically

Data loss prevention (DLP). Workspace rules can warn or block when sensitive details, like card numbers or health information, are shared from Gmail, Drive or Chat.
Chrome Enterprise Premium. Can stop sensitive information being pasted or uploaded into AI websites.
Sensitive Data Protection. Finds and labels sensitive data stored in Google Cloud.
Model Armor. Screens prompts and answers for AI apps you build on Google Cloud.

4. Govern your AI agents

AI agents can take actions on their own, so they need their own guardrails. In Google Cloud's Gemini Enterprise Agent Platform, the Govern section brings these together:

Agent Registry. One list of every AI agent your business runs. Agents deployed on the platform are added automatically.
Policies. Rules for what each agent may reach, and what it may never touch.
Gateways. A checkpoint for agent traffic, in and out, that enforces your policies and can screen content with Model Armor.
Security. A view of security settings and findings for your agents.
Topology. A visual map of which tools and services each agent talks to, so you can spot anything unexpected.

Not on Google?

Other platforms have similar tools. Microsoft Defender for Cloud Apps can discover generative AI apps in use, and Microsoft Purview can apply data protection to AI. Security tools called cloud access security brokers (CASBs) and secure web gateways can spot AI services from network traffic, whatever platform you use.

Keep the trust

Tell your team before you turn on monitoring, and explain why. Some states require employers to give notice first. Visibility works best alongside the conversation, not instead of it.

Some of these tools need a business or enterprise plan. Setting them up, and working out what your current plan already includes, is part of our AI governance work.

What this can look like

Example scenario. A three-person design studio asks everyone which AI tools they use for work. They find four, two of them on personal accounts. They choose one business plan, move their work into it, write a half-page policy, and set a calendar reminder to review it each quarter. The whole thing takes an afternoon.

The takeaway

Shadow AI is usually a sign that your team wants to work smarter. With a short list of approved tools, a simple policy and someone keeping an eye on it, you get the benefits of AI and keep your information where it belongs.