Shadow AI in small businesses: how to find it and what to do
Shadow AI in a small business is any AI tool staff use for work that nobody approved. The fix is four steps: find what is in use, approve a good alternative, redirect people to it, and record the result. Banning AI outright tends to push use out of sight. A managed list of approved tools, with client data kept out, works better and gives you something to show clients and insurers.
What shadow AI looks like
It is rarely dramatic. It is ordinary people trying to save time.
Take Ashcombe, a fictional 20-person accountancy practice. Priya drafts client letters in the free version of ChatGPT. Tom summarises meeting notes with an AI extension he found himself. Jo tried DeepSeek because a friend mentioned it. None of them asked. None of them meant harm.
That is shadow AI. Tools in use for work, chosen by individuals, with no one deciding what data may go in.
Why it happens
Staff use AI because it helps. Writing, summarising and checking all get faster. Most of it costs nothing to start.
A few common reasons it stays hidden:
- No one said which tools are allowed.
- The approved option is slower, or does not exist.
- People assume a free account is fine for work.
- A past ban made people stop mentioning it.
The last one matters. If asking feels risky, people stop asking. Use carries on. You just lose sight of it.
The real risks
AI is not the problem. Unmanaged data is.
The main risks for a small firm are plain:
- Client data leaves the firm. Names, bank details or a full client email get pasted into a tool you have no agreement with.
- Training and retention. Some free and personal accounts may keep what is typed, and settings differ by plan. Our guide on whether ChatGPT or Copilot trains on your business data covers this.
- Data protection duties. Under UK GDPR you need to know where personal data goes. You cannot answer that for tools you do not know about.
- Questions you cannot answer. Clients, auditors and cyber insurers increasingly ask about AI controls. "We are not sure" is a weak reply.
Step 1: find out which tools are in use
Start by asking. Keep it friendly. Make clear no one is in trouble.
A short survey works for small teams. Ask three things: which AI tools you use, what for, and what kind of data goes in. Then check what you can see yourself: browser extensions, sign-ups with work email addresses, and any expense claims for AI subscriptions.
If you have an IT provider, they may be able to see which AI sites are visited from work PCs. There is more detail in how to find out which AI tools your staff are using.
Rule of thumb: treat the first round of discovery as an amnesty. People who own up now are the ones who will ask first next time.
Step 2: approve a good alternative
People will only switch if the approved tool is as useful as the one they found. So pick one or two that cover the real tasks from your survey.
For most small firms that means a business plan of a mainstream tool, such as ChatGPT or Copilot, where the provider does not train on your data by default and offers a data processing agreement. Check the specific plan's terms before approving it. Settings differ between free, personal and business accounts.
Write down the decision. A simple table does the job.
| Tool | Status | Allowed for | Not allowed |
|---|---|---|---|
| ChatGPT (business plan) | Approved | Drafting, summarising | Client identifiers, bank details |
| Copilot (work account) | Approved | Email and document help | Passwords, secret keys |
| Free personal AI accounts | Not approved | Nothing work-related | All work data |
| DeepSeek | Not approved | Nothing work-related | All work data |
That table is the start of your approved list. Put it in your AI policy. Our UK AI policy template for employees has space for it.
Step 3: redirect, do not just block
Blocking an unapproved site with nothing in its place sends people to their phones. Redirecting works better.
Tell each person which approved tool replaces the one they used. Show them how to sign in. Then set out the data rules in plain words: no client names, no bank details, no National Insurance numbers, no passwords.
Rules on paper rely on memory. A busy afternoon is when a full client email gets pasted. So many firms add a check in the browser that catches sensitive data before it is sent. Our guide to stopping staff pasting client data without banning AI covers the options.
Step 4: record what you decided and who agreed
A record turns good intentions into something you can show. Keep:
- The approved list, with the date it was last reviewed.
- Each person's dated agreement to the AI policy.
- Who still needs a short training session.
- Any new tools that turn up, and what you decided about them.
Review the list every few months. New tools appear often. Your list should keep up.
AI is welcome here. Company and client data stay out of it.
Where Helio Guard fits
Helio Guard does steps 3 and 4 on Windows PCs, in Chrome and Edge. It checks every paste, message and file upload on listed AI sites. Sensitive items are taken out and swapped for labels, so the question still goes through. Client names can become a stand-in such as "Client A", then turn back when the answer is copied. Checked on the PC, with fixed rules. We never see what your team types.
It also flags unapproved AI tools, records dated policy agreement, shows who needs training, and produces a monthly record for clients, insurers or auditors. Windows only for now. Not yet on Mac.
It is £1 a person a month, from 5 people, with a 14-day free trial and no card needed. You can see the details on the pricing page.
Common questions about shadow AI
Should a small business just ban AI tools?
A ban on AI tools in a small business usually moves use out of sight rather than stopping it. Staff switch to phones or personal laptops, where you have no control at all. Approving one or two good tools, with clear data rules and a check before anything is sent, gives staff the help they want and keeps client data out.
Is using the free version of ChatGPT for work shadow AI?
Using the free version of ChatGPT for work counts as shadow AI if the business has not approved it. Free and personal accounts often have different data settings from business plans. If staff find the free version useful, the sensible response is to approve a business plan and move them across.
How often should the approved AI tools list be reviewed?
An approved AI tools list is worth reviewing every few months, and whenever someone asks to use a new tool. New AI products and features arrive often, and providers change their terms. A dated review also helps when a client, auditor or insurer asks how you manage AI use.
Related: How to monitor employee AI use without spying on staff