What to Do When Your Team Starts Using AI Tools You Didn't Approve
Shadow AI — staff using ChatGPT, Notion AI, or other tools without approval — is already happening in most small businesses. Ignoring it creates real data and compliance risk. Here's how to get ahead of it without killing the productivity gains.
Shadow AI is what happens when your team starts using AI tools you never sanctioned — ChatGPT for drafting proposals, Grammarly for client emails, Notion AI to summarise meeting notes — and nobody told you. It is already in most small businesses. The question is whether you find out now, on your terms, or later, after something goes wrong.
Why shadow AI spreads so fast in small teams
Small teams move quickly. When someone discovers a tool that saves them an hour a day, they use it. They don't file a change request. They don't ask IT — there often isn't an IT department. They just get on with it, because that's exactly the culture you've built: people who take initiative.
The tools are also genuinely easy to access. ChatGPT, Claude, Gemini, and dozens of specialised copilots are free or cheap, require no installation, and produce results that look good immediately. A customer service rep pastes in a client complaint and gets a polished reply in ten seconds. A sales manager feeds your pricing sheet into an AI to build a quote template. A bookkeeper uploads a PDF invoice batch to a tool that promises to extract the data automatically.
All of that sounds helpful. Some of it is. But each of those actions may have just sent your business data — client details, pricing, financial records — to a third-party server under terms your business never agreed to.
What the actual risks are
Business owners tend to react to shadow AI in one of two ways: panic and ban everything, or shrug and assume it's probably fine. Neither is right.
The risks worth taking seriously:
Data leaving your control. Most consumer AI tools train on user inputs by default unless you explicitly opt out or pay for an enterprise plan. A staff member pasting a client's financial data or personal contact details into a free AI tool may be breaching your obligations under the Australian Privacy Act or UK GDPR without realising it.
Inconsistent outputs reaching clients. If several people on your team are using different AI tools to draft client-facing content, and nobody is reviewing for accuracy, errors will eventually reach a client. AI tools hallucinate. That is not a flaw that will be patched away — it is structural to how large language models work.
Vendor lock-in and hidden costs. Free tools become paid tools. Staff build workflows around them. Then the price goes up, the tool pivots, or the free tier disappears — and you have a dependency you didn't plan for.
The risks that are often overstated: staff becoming lazier, or AI producing work that's obviously bad. Most small business AI output is fine for internal use. The problem is almost always about what data went in, not what came out.
How to find out what's already in use
Before you set any policy, you need a picture of what's actually happening. This doesn't need to be a formal audit. A short conversation — framed as curiosity, not interrogation — is usually enough.
Ask each team member two questions: "What tools do you use day-to-day that help you go faster?" and "Has anything new crept in over the last few months?" You'll get honest answers if you make clear you're not looking to punish initiative.
Document what comes back: tool name, what it's used for, whether it touches client data or financial data, and whether anyone is paying for it personally (which happens more than you'd expect).
In most small businesses I work with, that exercise surfaces between three and eight unapproved tools — and at least one is handling something sensitive.
What to do with what you find
Not everything on that list needs to be shut down. Some of it is worth formalising. Some needs to be replaced with a safer alternative. A small amount needs to stop immediately.
Keep and formalise — tools already providing clear value, handling only non-sensitive data (drafting internal comms, summarising public documents, generating ideas), with acceptable privacy terms under a paid plan.
Replace or upgrade — tools doing useful things with sensitive data, where a paid or enterprise version would give you the data processing agreements and opt-outs you need.
Stop now — tools where staff are inputting client PII, financial records, or confidential business data into free consumer tiers with no data protection agreements in place.
The goal is to make deliberate choices about which AI sits inside your control and which sits outside it. If you've already written an AI policy — or are thinking about it — this triage feeds directly into that process. There's more on structuring that in the post on how to write an AI policy for a small business.
How to respond to your team without killing the culture
The wrong move is to come back from this exercise with a blanket ban and a lecture about data security. You'll lose the trust of exactly the people who were trying to make the business better.
Be transparent instead. Tell the team what you found, what concerned you, and what you're doing about it. Most people don't know about default training data policies on free AI tools — that's not negligence, it's just not obvious. Explain it plainly and they'll usually understand immediately.
Then give them a path forward. If you're going to sanction specific tools, say so. If you want AI-assisted client work to go through a review step, explain why and keep the step lightweight. People accept constraints much more readily when they can see they're part of a plan rather than a clampdown.
For businesses ready to move from ad hoc AI use to something more structured, a fractional Chief AI Officer can set the strategy and guardrails without a full-time hire.
FAQ
Is shadow AI actually illegal? Using an unapproved tool is not itself illegal. But if that use leads to a privacy breach — sending personal data to a third party without a lawful basis — your business may be in breach of the Australian Privacy Act or UK GDPR. The tool isn't the problem; what data goes into it is.
What if a staff member refuses to stop using a tool they find valuable? Start by understanding why. If the tool is solving a real problem, find an approved alternative that solves the same problem. A flat "no" without a replacement just creates workarounds. If after that conversation the person still refuses, it becomes a conduct issue — but that's rarely where this ends up.
Do I need enterprise versions of every AI tool my team uses? No. Many tools are fine on free tiers for work that doesn't touch sensitive data. The upgrade decision comes down to one question: does this tool ever see client data, employee data, or confidential business information? If yes, check the data terms and upgrade or switch if they're not acceptable.
How do I stop this happening again with new tools? Add one lightweight step: before anyone brings a new tool into regular use, they flag it to whoever owns your AI policy. A five-minute check is all it takes. The goal is visibility, not bureaucracy.
General information, written to be useful — not financial, tax, investment or legal advice. For decisions specific to your business, take advice from a suitably qualified professional.
Think this might be a fit?
Tell us what you're trying to improve. We'll come back on whether it's a fit and a sensible next step — usually a short call and a free AI & Automation Health Check.