Shadow IT isn't new. I ran an IT services business for years before Lumitec, and staff quietly signing up for their own software was always part of the job. AI has made it faster and more serious, because these tools don't just store your data. They read it, write with it and increasingly act on it.
Here's what shadow AI is, how to find what's in use and what to do with it.
What is shadow AI?
Shadow AI is any AI tool, workflow or agent used for work without the business approving it. A free chatbot someone pastes customer emails into. A browser extension that summarises every page it sees. An AI assistant connected to your accounts system by someone who wanted a quicker report.
It's common. In Pax8's October 2026 survey of 1,150 businesses with 5 to 499 staff in the US, UK, Germany and Sweden, 35% of those using AI had found tools, workflows or agents running without approval in the previous six months. We covered the wider findings in this week's AI news.
What are the risks of shadow AI?
The main risk is data going somewhere you haven't checked. Respondents to the same survey named data security or privacy as the top risk of AI built by employees, in every country surveyed. The four we'd worry about most:
- Customer and staff data leaving the business. Free and personal plans can have different data terms from business plans. If nobody has read them, you don't know where that data went.
- Connections nobody owns. AI tools now plug straight into email, files and finance. Xero launched a connected app for Google Gemini in October, on top of its existing links to Microsoft 365, Claude and ChatGPT. Useful, and fine if you approved it. If nobody did, check who connected it and what it can see.
- Work nobody checked. A quote, a contract summary or a customer reply drafted by AI and sent without review.
- Bills nobody expected. 49% of the businesses paying for AI in Pax8's survey had a bill higher than expected. Usage-based charges were the hardest to forecast.
Should you ban AI tools at work?
Usually not. A ban can push use onto personal accounts and phones, which are harder to check. People use these tools because they save time. If the approved route is slower than the workaround, the workaround wins.
The better approach is to find what's in use, keep what's worth keeping on proper terms and give people an approved option that's genuinely easier.
How do you find the AI tools your staff already use?
Start by asking, then check. In that order, because asking first tells you why people chose each tool, and the reason matters as much as the tool.
- Ask every team, without blame. "Which AI tools do you use for work, and what for?" Make it clear nobody is in trouble. Asking first often turns up tools an account check misses.
- Check your Microsoft 365 or Google Workspace admin console. Both list third-party apps that staff have connected to their work accounts. Your IT provider can pull this for you.
- Check connected apps in your key systems. Accounts, CRM, ticketing and file storage. Look for anything AI-related nobody can explain.
- Check card statements and expense claims. AI subscriptions show up as small monthly charges on personal and company cards.
- Check your existing tools. AI features are being switched on inside software you already pay for. Some of your shadow AI arrived in an update.
What to do with what you find
Sort every tool into three groups:
- Keep and make official. It's useful and the terms are acceptable. Move it onto a business plan, give it an owner and add it to your approved list.
- Replace. The job is worth doing but the tool isn't safe for your data. Find an approved tool that does the same job.
- Stop. It touches data it shouldn't and there's no good case for it. Disconnect it and explain why.
Then write a short AI policy. One page is enough: which tools are approved, what data must never go into AI, who checks AI output before a customer sees it and who to ask before trying something new. Review it every quarter, because the tools change that often.
How we approach it with clients
The answer to shadow AI is usually a better approved option, not a stricter rule. That's how we design what we build.
Take the bid-response engine we built for a franchise bid team. The first phase is live and in use. It drafts answers in the business's own tone from more than ten years of its real bids, winning and losing. The team gets the speed they'd want from a chatbot, in a tool built for the job, with every answer traced back to real bid material. It wasn't a shadow AI clean-up. It's the decision we'd make in one: give people an approved tool grounded in the business's own material.
The same thinking runs through the AI service desk we're building for a US managed service provider, which is in delivery now. It triages tickets against rules the client agreed, with tight limits on what it can change. Every decision about what it can see and do was made on purpose.
Checklist: how to check for unapproved AI tools
- Every team has been asked which AI tools they use.
- Connected apps in your Microsoft 365 or Google Workspace admin console have been reviewed.
- Connected apps in your accounts, CRM and file systems have been checked.
- Card statements and expenses have been checked for AI subscriptions.
- Each tool is marked keep, replace or stop.
- Every approved tool has a named owner and a spend cap where one exists.
- A one-page AI policy is written, shared and dated for review.
Where we fit
Your IT provider can help check connected apps, and asking teams takes a little longer. The harder part is deciding which jobs are worth doing with AI and building an approved way to do them. That's what we do. We build AI with access and spend controls designed in, then run it so it doesn't drift.
The AI Readiness Diagnostic is free and no-obligation. We look at how you work, pick the jobs worth doing with AI and hand you a statement of work, a wireframe and an ROI case you keep whether or not we build anything. If you're an IT provider whose clients are asking about this, our AI for MSPs page shows how we work alongside you.
Common questions
Is it a data breach if staff put customer data into ChatGPT?
Not automatically, but it can be. It depends on what data it was, which account and plan were used, the terms behind them and what happened to the data. Personal accounts are the bigger worry because the business hasn't checked their terms, but business accounts need checking too. Decide which tools are approved for personal data and say so in writing.
Do small businesses need an AI policy?
Yes, if anyone in the business uses AI for work, and most do. It doesn't need to be long. A single page covering approved tools, data that must never go into AI, who checks AI output and who to ask before trying something new does the job. Review it every quarter.
Can my IT provider tell which AI tools staff are using?
Partly. They can see apps connected to work accounts in Microsoft 365 or Google Workspace and, depending on your setup, some web traffic. They can't see personal accounts on personal phones. That's why asking staff directly, without blame, is the first step rather than the last.
See it in practice: AI bid-response engine for a franchise bid team →