What to Do When an Employee Pastes Client Data Into an AI Tool

When an employee tells you they pasted client information into an AI tool, treat it as a security incident, not a disciplinary meeting. Put one person in charge. Write down what was shared before anyone deletes anything. Stop further sharing, change any passwords or keys that were in it, and rank how sensitive the data was. Then decide with your lawyer and insurer whether anyone has to be told.

It happens more often than most owners think. In a survey DNSFilter released on November 3, 2026, 46.1% of organizations said an employee had connected an unapproved AI tool to Google Workspace, Microsoft 365, or their single sign-on environment, and 20.1% said one of those connections caused a security incident. A single paste is usually the one you heard about, not the only one.

First, don't delete anything

The natural reaction is to tell the employee to delete the chat. Don't, at least not yet. That conversation is your best record of exactly what left the business, when, and through which account. Delete it before you've written that down and you lose your evidence, and you'll have a harder time answering your lawyer, your insurer, or your client.

Deleting a chat is cleanup. It doesn't take back anything already shared.

The first hour: six steps

Name one person in charge. One owner for the response, usually whoever handles IT or compliance. Keep the tone blame-free. You want the next person to come forward just as quickly.

Write down the facts. Which tool. Which account: free, personal, paid, or a company-managed business account. The date and time. Exactly what was pasted or uploaded. Whether any share links, projects, or connected apps were involved.

Stop further sharing. Pause the account, or disconnect the browser extension or app connection that's involved. Disable rather than uninstall where you can, so the logs survive.

Change any secrets right away. If a password, API key, or access token was in the paste, revoke it and issue a new one now. That holds whatever plan the employee was on.

Check sign-ins and app permissions. In Microsoft 365 or Google Workspace, review recent sign-ins and the third-party apps that have been granted access to mail and files. Look for AI tools nobody approved.

Save a record, then ask for deletion. Take screenshots or an export of the conversation for your file. Then request deletion from the vendor and keep proof that you asked.

How bad is it? Rank what leaked

Not every paste carries the same risk. Rank it so your response matches the problem:

1. Secrets and credentials: passwords, API keys, access tokens. Highest urgency, because someone can use them directly. Rotate first, investigate second.
2. Regulated personal data: Social Security numbers, health information, payment card or bank details. This is where notice obligations most often come into play.
3. Client-confidential material: contracts, financials, or work product covered by an NDA or client agreement.
4. Internal material: drafts, internal notes, general business information. Lower risk, but still worth recording.

The account type matters as much as the data. Free and personal AI accounts and company-managed business accounts usually come with different retention, training, and admin terms. Check the current terms for the specific tool and plan involved. They change, and the answer depends on the version your employee used.

Do you have to tell anyone?

Maybe. This is a legal call, not a technical one, and it isn't something to decide alone. Depending on the data and your industry, the obligations that may apply include:

• Your client contracts and NDAs, many of which set their own notice terms
• The FTC Safeguards Rule, for firms handling customer financial information, including many accounting and tax practices
• HIPAA, for healthcare practices and their business associates
• State breach-notification laws, including the Florida Information Protection Act
• Your cyber insurance policy, which may require prompt notice of a suspected incident

Your job in the first day is to have the facts ready: what was shared, through which account, and what you've already done. Counsel decides whether notice is required. This section is general information, not legal advice.

Talking to the employee, and the client

Coach, don't punish. Most people who paste client data into an AI tool are trying to get work done faster, and nobody told them where the line was. If the first person to admit a mistake gets disciplined, the next one won't admit it at all, and you'll find out months later.

If a client needs to hear about it, be direct and brief: what happened, what data was involved, what you've already done to contain it, and what you're changing so it doesn't happen again. Have counsel review the wording first.

The bigger problem: tools you don't know about

One paste is usually a symptom. If one employee reached for an unapproved tool, others probably have too, and some of those tools may have standing access to your email and files through app permissions granted months ago. That's shadow AI, and finding it is the step that turns one incident into a lasting fix. Our plain-English guide to shadow AI explains how businesses find out what's already in use.

How to stop it happening again, without banning AI

Banning AI doesn't stop AI use. It moves it onto personal phones where you have no visibility at all. What works is a short list of controls:

• A one-page AI acceptable-use policy with a plain data line: what can go into an AI tool, what needs approval, and what never goes in.
• An approved business AI tool with single sign-on, logging, and terms that keep your content out of model training, so staff have a safe option that's also the easy one.
• Admin approval for AI app connections to email, calendars, and file storage, instead of letting anyone grant access.
• Monitoring of which AI services are being reached from company devices and accounts.
• A redaction habit: replace names and identifiers with placeholders like [CLIENT_A] before anything goes into a tool that isn't approved for that data.

If you're starting from nothing, the free companion worksheets for Securely Integrated AI, Jason Radney's book, include templates for AI acceptable use, shadow AI discovery, data classification, and incident response. Download the worksheets.

Straight answers

Is pasting client data into an AI tool a data breach?

It can be. That depends on what was shared, whose data it was, the account's retention and training terms, and the laws and contracts that cover that data. Social Security numbers pasted into a free personal account are far more serious than a generic draft in a company-managed business account. Record the facts first, then let counsel make the legal call.

Does deleting the conversation fix it?

No. Deleting the chat is cleanup, not containment. Record the details first, revoke and replace any credentials that were included, then ask the vendor to delete the data and keep proof of the request.

Should we discipline the employee?

In most cases, no. Treat the first incident as a training gap and fix the policy and tools. People keep reporting mistakes when reporting them is safe.

How SecurelyIntegrated.AI can help

We help small and mid-sized companies across Tampa Bay and nationwide find the AI tools already in use, draw a clear data line, and configure approved tools so staff stop reaching for risky ones. The AI Readiness Assessment is two weeks and fixed fee, and it delivers a full inventory of AI tools in use, a data exposure map, a risk register, and a draft acceptable-use policy ready to adopt. If you'd rather take a quick self-check first, the free 10-question AI readiness scorecard at HelpUsAdopt.AI takes about five minutes.