
Most teams are using AI already, often on personal accounts with client details in the prompt. Here is how to find it without turning the office into an investigation.
Ask most business owners whether their team uses AI at work, and the answer is usually no. Then ask whether anyone has a personal chat account open in another browser tab.
That gap between the answer and the truth is what people mean by shadow AI. It is rarely someone breaking a rule on purpose. It is someone with a stack of notes to summarize and no other way to get home before seven.
The tool is not the problem. Where the client’s information ends up is the problem.
Why it happens
Three things push people toward tools nobody approved.
The work repeats. Summarizing intake notes, drafting the same follow-up email, pulling numbers out of a PDF. When a task is boring and has an obvious shortcut, people take it.
The approved route is slower. If asking for help means a ticket and a two-week wait, using a personal account takes forty seconds.
Nobody said what was allowed. Silence feels like permission, especially when a client is waiting.
What it actually costs
When someone pastes a client’s medical history, bank details, or contract terms into a personal account, that information has left your building. Depending on your industry, that may count as a reportable disclosure. Even when it does not, you have no record of what was shared, no way to delete it, and no chance to catch a wrong answer before the client reads it.
Worth knowing: on many free and personal plans, conversations can be used to improve the model unless the person turns that off in settings. A business plan with a signed data agreement is a different arrangement. Most people using a free account have no idea which one they are on.
How to find it without an inquisition
Do not announce an audit. You will get clean answers and no information. Start with a normal conversation instead.
Ask what is slowing them down. “What part of your week feels like copy and paste?” You will get a list, and it will match the shortcuts they are already taking.
Ask what they have tried. Say plainly that the answer cannot get anyone in trouble, then mean it. The first person who admits it decides whether anyone else talks.
Look at what leaves the building. Attachments, shared drives, browser history on company devices. You are looking for patterns, not for someone to blame.
Then say it out loud
People do not want to break rules. They want to know the rules. Two short lists handle most of it.
- Allowed: the tools your team may use, and what kind of work each one is for.
- Never: the information that stays out of every tool. Client names attached to health or financial details, anything under an NDA, and anything you would not put in an email to a stranger.
Add one line about being unsure. Ask first, no penalty. That line does more for adoption than the rest of the policy.
Close the gap with access
Rules alone do not fix this. Give the team a route that is as fast as the shortcut they were using: an approved account, a workspace where history is controlled, and a few saved templates for the tasks they do every day.
When the approved way is also the easy way, the personal accounts quietly stop being opened. That is the whole job. Find it, name it, replace it with something better.
