Ask a small business which AI tools its staff use and you’ll usually hear one answer: “We gave everyone Copilot.” Then look, and you find ChatGPT on personal accounts, two meeting note-takers, a browser extension that summarises every page, and an AI app connected to someone’s mailbox in March and never used again.
That’s shadow AI: AI tools in use at work that nobody approved, registered or is watching. It’s the fastest-growing kind of shadow IT. You can’t govern what you can’t see, so detection is the first step. Here’s where shadow AI hides and how to find it.
How common is shadow AI?
- 78% of people who use AI at work bring their own AI tools, and the figure is 80% at small and medium-sized companies. (Microsoft and LinkedIn 2024 Work Trend Index)
- One in five organizations in IBM’s study had a breach involving shadow AI, and a high level of shadow AI added an average of $670,000 to breach costs. (IBM Cost of a Data Breach Report 2025)
- 63% of breached organizations had no AI governance policy or were still writing one. (IBM Cost of a Data Breach Report 2025) If that’s you, start with our one-page AI usage policy template.
Banning AI doesn’t make it go away. It just moves it to personal accounts and phones, where you have even less visibility.
Where shadow AI hides
Shadow AI isn’t one thing. It shows up in five places, and each needs a different way to look:
- AI websites in the browser. ChatGPT, Claude, Gemini, Perplexity and hundreds of smaller tools, often signed in with a personal account.
- AI apps connected to mail and files. Note-takers, schedulers and “AI assistants” that people approve with two clicks, granting access to their mailbox, calendar or drive.
- Browser extensions. “AI summary” and “AI writing” extensions that can often read every page the person opens.
- AI built into tools you already pay for. Copilots and assistants switched on inside SaaS apps, sometimes by default.
- Desktop apps and AI agents. AI apps installed on laptops, and agents that act inside the browser on someone’s behalf.
Five ways to detect shadow AI
1. Review app consents in Microsoft 365 and Google Workspace
Both platforms record every third-party app a user has granted access to. In Microsoft Entra ID it’s under Enterprise applications. In the Google Workspace admin console it’s under Security, in API controls, where third-party app access is managed. Look for AI note-takers, assistants and anything with access to mail, calendar or files. Note who approved each one and when it was last used.
This catches the riskiest kind of shadow AI, because these apps keep their access long after anyone remembers them.
2. Look at browser and network traffic for AI domains
If you run DNS filtering, a secure web gateway or a firewall with web categories, filter for AI services. You’ll see which tools are visited and how often. It won’t tell you whether staff used a work or a personal account, or what they pasted.
3. Inventory browser extensions
Chrome and Edge can report installed extensions across managed browsers. Search the list for AI, GPT, “summary”, “writer” and “assistant”, and check what permissions each one asks for. “Read and change all your data on all websites” is the one to worry about.
4. Check what’s installed on laptops
Your RMM or endpoint tool can list installed software. Look for desktop AI apps and anything that installs its own browser or agent.
5. Ask people, without blame
A short, anonymous survey often finds tools the logs miss, especially on phones. Ask what they use and what for, and say plainly that the goal is to find safe, approved tools, not to punish anyone.
A worked example: one week of looking
This is an illustrative example, not a real client. An MSP runs these checks for a 40-person architecture firm that “only uses Copilot”.
- App consents: 14 AI apps connected to staff mailboxes or drives. Three are meeting note-takers with full calendar and mailbox access, and one hasn’t been used since spring.
- Web traffic: 11 different AI sites visited in a week. ChatGPT is the busiest, mostly from personal accounts.
- Extensions: 6 AI extensions. Two can read every page the user opens.
- Survey: two project leads use an AI image tool on their phones for client concept work.
None of it was malicious. But the firm’s client drawings and contracts were reaching at least a dozen services nobody had reviewed.
What to do with what you find
Detection only helps if it leads somewhere. Work through what you found in this order:
- Revoke what’s unused. Old AI app grants with mailbox or file access are the quickest win. Nobody will miss them.
- Approve a good default. Pick one approved AI tool for general work, so people have somewhere safe to go.
- Write it down. Keep a register: each AI tool, who uses it, what it can reach, who owns it, and when it’s next reviewed.
- Set a simple rule for data. For example, no client data in personal AI accounts.
- Coach, don’t just block. When someone hits a rule, tell them why and point them to the approved tool.
- Look again monthly. New AI tools appear every week. Detection has to be continuous, not a one-off project.
Shadow AI detection checklist
- App consents in Microsoft 365 or Google Workspace reviewed for AI apps.
- Unused AI app grants with mail or file access revoked.
- AI websites in use identified, including personal-account sign-ins.
- Browser extensions inventoried and risky ones removed.
- Desktop AI apps on laptops listed.
- One approved AI tool chosen and announced.
- An AI register with an owner and review date for each tool.
- A clear rule on client and company data in AI tools.
- The checks repeated every month.
How Centeye helps
Centeye Lumen is the AI governance and registry platform. Connect Microsoft 365 or Google Workspace in under 10 minutes and it lists every connected AI app and what each one can reach. Add the Chrome and Edge extension and the Windows sensor, and it also sees which AI tools are in use and by whom, including personal accounts, and how much data is pasted or uploaded. It never reads the content.
Lumen keeps the register for you, with owners and review dates, revokes forgotten access at your setting, blocks risky extensions, and coaches people in the moment with a short note pointing to the approved tool. Every revoke is checked at Microsoft or Google and can be undone.
New to the topic? Read what shadow AI is, how company data leaks into AI tools, or browse the security glossary. Want to see what’s in use at your own company? Start a free trial.
Frequently asked questions
What is shadow AI?
Shadow AI is any AI tool used for work that the company hasn’t approved or isn’t tracking: AI websites, browser extensions, apps connected to mail and files, or AI features switched on inside other software.
How do I find out which AI tools employees are using?
Start with app consents in Microsoft 365 or Google Workspace, which show AI apps with access to company data. Then look at web traffic for AI sites, inventory browser extensions, and check installed software. A short, blame-free survey fills the gaps.
Should we just block AI tools?
Blocking everything usually pushes people to personal accounts and phones, where you can’t see anything. It works better to approve a good default tool, block the risky ones, and coach people when they go off-track.
Is shadow AI only a problem for large companies?
No. Microsoft found that bringing your own AI tools is slightly more common at small and medium-sized companies than in general. (Microsoft and LinkedIn 2024 Work Trend Index)
Figures cited from the Microsoft and LinkedIn 2024 Work Trend Index and the IBM Cost of a Data Breach Report 2025. The worked example is illustrative. Centeye describes only capabilities that ship in the product today.