Intercept IT All articles
Cybersecurity

The Browser Tab You Didn't Approve: A Plain-English AI Data Audit

Intercept IT
The Browser Tab You Didn't Approve: A Plain-English AI Data Audit

It usually starts with good intentions. Someone has a client call, pastes the transcript into a free summariser, and gets a tidy set of notes in seconds. Another person uses a browser-based assistant to polish a quote, or a spreadsheet add-on to clean up a contact list. Nobody is trying to cause trouble. They are trying to get home on time.

The trouble is that the client data goes along for the ride. Names, email addresses, pricing, complaint details, maybe a contract clause or two. Not every free AI tool is a disaster, but if nobody in the business knows what is being used, you cannot make a sensible call about risk. That is the gap a shadow-AI audit closes.

This is not a ban on AI

A blanket ban rarely works. Staff will use the tools anyway, just less visibly. The useful question is not whether people use AI at work. It is which tools are touching which data, and whether that fits how you already handle client information.

For a UK SME, the aim is simple: know what is in use, remove the obviously risky habits, and give everyone a short rule they can follow without asking a lawyer for permission to send an email.

Start where the work actually happens

You do not need forensic software for the first pass. Ask people to show you their browser bookmarks and extensions, not their whole browsing history. Most shadow AI shows up in plain sight: a meeting recorder that joins calls, a writing assistant in a sidebar, a PDF summariser, an image generator, a spreadsheet plug-in.

Look for tools that ask staff to paste text into a box. That is the common pattern. If a client proposal, a support ticket or a staff rota is being pasted into a free web tool, the data has left your systems. The tool may be perfectly legitimate, but you still need to know it exists.

Also check the add-ons inside Microsoft 365, Google Workspace and your browser. A small extension can read the page you are on, which is fine for a grammar checker and less fine for something that hoovers up client records. Remove anything nobody can explain.

Check what your existing accounts already expose

Here is the bit many SMEs miss. You may not need a new AI tool to have an AI data problem. If your Microsoft 365 or Google Workspace permissions are too loose, an approved tool can reach more than it needs.

In Microsoft 365, look at which third-party apps have been granted access by users, and review sharing links on sensitive folders. In Google Workspace, check third-party app access and whether staff can install marketplace add-ons without approval. You are not trying to lock everything down for the sake of it. You are checking whether a free tool can read a client folder because someone clicked allow in 2024 and nobody looked again.

A simple test: if a new starter joined tomorrow, could they find client files through a shared link that was never meant to be public? If yes, fix that before you worry about the latest AI app.

The tea-break audit

Set aside thirty minutes and do this with one person from operations, not just IT. Ask three questions.

First, which AI tools do you use for work, even if you pay for them yourself? Write them down without judgement. Second, what kind of information do you put into them? Client names, financial figures, health details, staff issues, legal advice. Third, does the tool need that information to do the job?

Then sort the list into three piles. Green means the tool is approved, the data is low-risk, and the settings are sensible. Amber means the tool might be useful but needs a proper look. Red means stop using it for client data today, while you check the contract and settings.

You will probably find that most tools are amber. That is normal. The point is to move from unknown to known.

Write a one-page rule, not a policy novel

A useful AI rule for a small business can fit on one side of A4. It should say which tools are approved, what data must never be pasted into a free tool, and who to ask when someone wants to try something new. Add a line about meeting recordings: get consent before recording a client call, and do not let a bot join without telling people.

Keep the language human. Something like: use approved tools for approved work; do not paste client, staff or financial data into free AI websites; ask before connecting a new tool to your work account. That is enough to change behaviour.

Make it a habit, not a panic

Revisit the list when someone joins, leaves or asks for a new tool. Check app permissions every quarter, alongside your usual password and backup checks. If a client asks how you handle their data, you will have a straight answer rather than a shrug.

The browser tab nobody approved is not a scandal waiting to happen. It is a sign that your team is trying to work efficiently. At Intercept IT, we would rather help you spot the useful tools and fence off the risky ones than pretend AI is not already in the building. A short audit now is cheaper than a awkward client conversation later.

All Articles

Related Articles

What Your Cyber Insurer Is Really Asking When It Says 'Do You Have MFA?'

What Your Cyber Insurer Is Really Asking When It Says 'Do You Have MFA?'

The Auto-Reply That Hands Over the Org Chart

The Auto-Reply That Hands Over the Org Chart

Who Still Has the Keys? The Leavers' Login Audit Your SME Probably Skipped

Who Still Has the Keys? The Leavers' Login Audit Your SME Probably Skipped