# What to Do If an Employee Pastes Customer Data Into ChatGPT

Canonical URL: https://business.theaieducator.io/posts/employee-pasted-customer-data-into-chatgpt
Publication: Dan Fitzpatrick Insights
Author: Dan Fitzpatrick
Topic: AI Governance and Risk
Published: 2026-11-10T16:40:57.000Z
Modified: 2026-11-10T16:41:06.479Z

An employee pasted customer data into ChatGPT? Use the five-step Paste Plan and a copyable incident log to stop it, scope it, decide who to tell and fix the cause.

## In brief

Treat a paste of customer data into an AI tool as a possible breach. Stop it, capture evidence, ask the provider about deletion, decide who must be told, then fix the cause. In the UK, reportable breaches go to the ICO within 72 hours of discovery.

## Key takeaways

- A BlackFog-commissioned survey published 4 November 2026 found 47% of security leaders at UK and US companies with 500+ staff had seen sensitive data exposed through an AI tool.
- Among those organizations, 77% said a company-approved AI tool was involved, so approving a tool does not replace a rule on what goes into it.
- The Dutch Data Protection Authority reported on 6 August 2024 that employees sharing personal data with AI chatbots had triggered several breach notifications.
- Capture the chat and the account details before you delete anything, then ask the provider about deletion.
- In the UK, a reportable personal data breach must reach the ICO without undue delay and within 72 hours of discovery, even if details are incomplete.
- In the US, all states have breach notification laws, so check state rules and talk to counsel; health data adds federal rules.
- The Paste Plan is five steps: Stop, Scope, Pull back, Decide, Fix. Keep a Paste Incident Log from minute one.

Treat it as a possible data breach and act within the hour. Stop more pasting, record what went in and which tool and account it went into, ask the provider about deletion, decide whether customers or a regulator must be told, then fix the cause. In the UK, a reportable breach must go to the ICO within 72 hours of discovery. The Paste Plan below sets the order.

## How common is this, and is a paste really a data breach?

It is common enough that at least one European regulator has warned about it, and in some cases, yes, it is a breach. A survey commissioned by the security vendor BlackFog and fielded by Sapio Research in September 2026 reached 400 IT security decision makers at UK and US companies with more than 500 employees. BlackFog published the results on 4 November 2026: 47% said their organization had experienced an incident in which sensitive data was exposed through an AI tool, and 77% of those said a company-approved AI tool was involved. The survey covers large firms, so it tells you nothing precise about a 20-person business. It does show that the problem is not limited to staff using tools you banned.

The regulator view is older and more specific. On 6 August 2024 the Dutch Data Protection Authority (Autoriteit Persoonsgegevens) said it had received several breach notifications caused by employees sharing personal data with an AI chatbot. Two examples: an employee of a medical practice entered patient data against the employer's instructions, and a telecom employee entered a file of customer addresses. I could not open the authority's own page for this piece, so I read the account in a law firm's summary (Alston & Bird, 9 August 2024) and label it that way in the sources.

Whether a given paste is legally a breach depends on your law, the data and the tool. Under the GDPR definition, a personal data breach includes unauthorized disclosure of personal data. Sending it to a third party you have no agreement with can fit that. A paste into a business account with a data processing agreement fits less clearly. Decide that after the facts are in your log, not before. Do not argue yourself out of acting in the first hour.

## What is the Paste Plan?

The Paste Plan is my five-step response to customer data pasted into an AI tool: Stop the pasting, Scope what went in, Pull it back where you can, Decide who must be told, and Fix the cause. The first three happen in the first hour. The last two can take a few days. It is my own suggested method, built on the UK Information Commissioner's Office (ICO) guidance for small organizations and the US Federal Trade Commission (FTC) breach guide. It is not a legal standard.

1. **Stop.** Tell the person to stop using that tool for work data, and do not paste anything else to "see what happened". Ask who else has used the same tool or account this week.
2. **Scope.** Capture the evidence before anything is deleted: a screenshot or export of the chat, the account name, the tool, the date, and what the paste contained. Count the people affected and note which fields were included, such as names, emails, addresses, account numbers or health details.
3. **Pull back.** Delete the chat only after you have captured it. Check whether the account lets you switch off training on your data and remove history. Contact the provider through its published route to ask about deletion. I have not checked any provider's current retention terms for this piece, so read the terms for the account the person actually used. The ICO's guidance says to recover or contain the data wherever you can, and to protect the people most affected.
4. **Decide.** Work out the risk of harm to the customers, then who must be told: the customers, a regulator, your insurer, a client whose contract requires notice. Write down the decision and the reason.
5. **Fix.** Find out why it happened. In most cases there is a gap: no approved tool, no rule on what must not be pasted, or no time to do the task another way. Close the gap so the next paste goes somewhere safe.

## What goes in a Paste Incident Log?

The log is the artifact that keeps you calm and protects you later. The ICO's small-business guide says to start a log even if you end up not reporting, recording what happened, who is involved and what you are doing about it. Copy the table below into a document and fill it in as the facts arrive.

| Field | What to write | Example (hypothetical) |
|---|---|---|
| Discovered | Date, time and how you found out | Tuesday 10:15, the employee mentioned it in a team chat |
| Who and what tool | Employee, tool, account type | Office administrator, free chatbot, personal login |
| What was pasted | Data fields and rough volume | 120 tenant rows: names, unit numbers, phone numbers, arrears amounts |
| People affected | Count and who they are | 120 tenants of one property |
| Stopped | Time pasting stopped, who confirmed | Tuesday 10:30, manager confirmed |
| Evidence kept | Screenshot or export, where saved | Chat exported to the incident folder at 10:40 |
| Provider action | Deletion request sent, settings changed | History off, deletion request sent Tuesday 11:20 |
| Risk of harm | Your judgment and your reasoning | Low to medium: contact details plus debt status, no financial logins |
| Who is told | Customers, regulator, insurer, client | Insurer notified; decision on regulator logged Wednesday |
| Cause and fix | Why it happened, what changes | No approved tool for letters; approved tool and written rule by Friday |

Take a hypothetical 15-person property management firm. An office administrator pastes a 120-row spreadsheet of tenant arrears into a free chatbot on her personal login to draft reminder letters. She mentions it on Tuesday morning. The manager has the log open within 15 minutes, and the clock the ICO describes, running from discovery, is already ticking. By 11:30 the paste has been stopped, the chat exported, history switched off and a deletion request sent. The rest of the table is a decision list, not a panic list.

## Who has to be told, and how fast?

In the UK, if the breach meets the threshold for reporting, the ICO says you must report it without undue delay and within 72 hours of becoming aware of it. The clock starts when you discover the breach, not when it happened. If there is a high risk to the people affected, you must also tell them without undue delay. You can report with partial details and follow up later. The ICO runs a self-assessment tool and an advice line for small organizations that are unsure whether a breach is reportable. Check the ICO's current pages, because UK data protection law was amended by the Data (Use and Access) Act 2025 and guidance is being updated.

In the US there is no single rule. The FTC's guide for businesses (published April 2019, modified 20 June 2025) says all states, the District of Columbia, Puerto Rico and the Virgin Islands have laws requiring notification of security breaches involving personal information, and that other laws may apply depending on the data. It tells you to talk to legal counsel. I did not check individual state laws for this piece, and health data adds federal rules such as HIPAA. If the data was health-related, treat that as the moment to call a lawyer.

Two other parties are easy to forget. Read your customer contracts, because some require you to tell a client if their data is handled outside agreed systems. And check your insurance policy wording for incident notice conditions. For more on how cover treats AI mistakes, see [whether business insurance covers AI mistakes](https://business.theaieducator.io/posts/business-insurance-cover-ai-mistakes). This article is general information, not legal advice. For anything involving health data, children's data or a large number of people, get qualified advice.

## What do you say to the employee?

Say thank you first, because the alternative is that the next person hides it. The Dutch authority's guidance stresses clear agreements between employers and employees about AI chatbots, with no ambiguity about what is allowed. If your agreement was never clear, the employee is partly reporting your gap. Ask three questions: what did you paste, which account did you use, and what were you trying to get done. The third answer tells you what the real fix is.

Do not turn a first incident into a disciplinary case unless there was clear bad faith or a repeat of a rule you had stated in writing. A culture of blame is a reporting problem. My view is that someone who pastes customer data into a free tool is usually trying to get through a pile of dull work faster. That points to a workflow problem before it points to a loyalty problem.

## What do I tell business owners who have just found a paste?

Do the first hour well and the rest gets easier. There are two easy mistakes: deleting the evidence in a hurry, and waiting to be certain before you start a log. Capture first. Log from minute one.

Then look at the larger pattern. BlackFog's survey found that among the organizations reporting an exposure, 77% said an approved tool was involved, so choosing the right tool does not remove the need for a rule on what goes into it. If this paste was one of several, you may have a [shadow AI problem](https://business.theaieducator.io/posts/shadow-ai-small-business) to deal with, and the next step is a short written rule. [Does a small business need an AI policy?](https://business.theaieducator.io/posts/does-a-small-business-need-an-ai-policy) covers the case for one, and the [UK small business AI policy template](https://business.theaieducator.io/posts/uk-small-business-ai-policy-template) gives you a starting point. A policy only works if people know it exists, so pair it with [AI training for small business teams](https://business.theaieducator.io/posts/ai-training-for-small-business-teams) and teach staff how to [check AI output](https://business.theaieducator.io/posts/train-staff-to-check-ai-output) before it goes anywhere.

## What should you do this week?

1. **Today:** copy the Paste Incident Log into a shared folder, so it exists before you need it.
2. **Tomorrow:** write one sentence everyone can follow, such as "No customer names, contact details or account numbers go into any AI tool unless the owner has approved that tool in writing."
3. **This week:** name one approved tool for the dull jobs that cause pastes, and tell the team which one it is.
4. **This week:** name who runs the Paste Plan when something happens, and who is the backup.
5. **Within a month:** run a ten-minute drill with the log and a made-up example, so the first real one is not the first time.

If you want to see how ready the whole business is, [is your business data ready for AI?](https://business.theaieducator.io/posts/is-your-business-data-ready-for-ai) is the next read.

## Where to go next

If your team is using AI but nobody has agreed the rules, that is what my [AI training for teams](https://theaieducator.io/ai-training-for-business?utm_source=business.theaieducator.io&utm_medium=referral&utm_campaign=employee-pasted-customer-data-into-chatgpt) is built for. If you would rather see where you stand first, take the [free Business AI Readiness Check](https://theaieducator.io/business-ai-readiness?utm_source=business.theaieducator.io&utm_medium=referral&utm_campaign=employee-pasted-customer-data-into-chatgpt). And if you are planning an event and want your members or leadership team to hear this argument in person, I give [keynotes to business audiences](https://theaieducator.io/business-keynotes?utm_source=business.theaieducator.io&utm_medium=referral&utm_campaign=employee-pasted-customer-data-into-chatgpt).

## Sources and further reading

- BlackFog, [Research: 47% of security leaders have experienced sensitive data exposure through AI tools](https://www.businesswire.com/news/home/20261104195397/en/BlackFog-Research-47-of-Security-Leaders-Have-Experienced-Sensitive-Data-Exposure-Through-AI-Tools), Business Wire, 4 November 2026. Vendor-commissioned survey by Sapio Research, September 2026, 400 respondents at companies with more than 500 employees in the UK and US.
- Alston & Bird, [Dutch Data Protection Authority warns that using AI chatbots can lead to personal data breaches](https://www.alstonprivacy.com/dutch-data-protection-authority-warns-that-using-ai-chatbots-can-lead-to-personal-data-breaches/), 9 August 2024 (summary of the authority's guidance of 6 August 2024).
- Information Commissioner's Office, [72 hours: how to respond to a personal data breach](https://ico.org.uk/for-organisations/advice-for-small-organisations/personal-data-breaches/72-hours-how-to-respond-to-a-personal-data-breach/), read 10 November 2026.
- Federal Trade Commission, [Data Breach Response: A Guide for Business](https://www.ftc.gov/business-guidance/resources/data-breach-response-guide-business), published 29 April 2019, modified 20 June 2025.
- On this site: [Shadow AI in a small business](https://business.theaieducator.io/posts/shadow-ai-small-business), [Does a small business need an AI policy?](https://business.theaieducator.io/posts/does-a-small-business-need-an-ai-policy), [UK small business AI policy template](https://business.theaieducator.io/posts/uk-small-business-ai-policy-template), [Does business insurance cover AI mistakes?](https://business.theaieducator.io/posts/business-insurance-cover-ai-mistakes), [AI training for small business teams](https://business.theaieducator.io/posts/ai-training-for-small-business-teams), [Is your business data ready for AI?](https://business.theaieducator.io/posts/is-your-business-data-ready-for-ai).

*Dan Fitzpatrick is [The AI Educator](https://theaieducator.io/about), a Forbes contributor, keynote speaker and bestselling author who has trained more than 150,000 people across 30+ countries.*

## Frequently asked questions

### What should you do if an employee pastes customer data into ChatGPT?

Stop further pasting, capture the chat and account details, then ask the provider about deletion. Work out the risk to the customers, decide who must be told, and fix the cause. In the UK, a reportable breach must go to the ICO within 72 hours of discovery.

### Is pasting customer data into ChatGPT a data breach?

Sometimes. Under the GDPR definition a personal data breach includes unauthorized disclosure of personal data. The Dutch regulator has received breach notifications for this exact behavior. Whether yours qualifies depends on the data, the tool, the account type and your law, so log the facts first.

### Do you have to report an employee pasting customer data into an AI tool?

It depends. UK businesses must report a breach to the ICO within 72 hours if it meets the reporting threshold, and tell affected people without undue delay if the risk is high. US rules vary by state. Contracts and insurance may also require notice. Take legal advice for health data.

### Should you delete the ChatGPT conversation straight away?

No. Capture a screenshot or export first, with the account name, tool and date, so you have a record of what went in. Then delete the chat, switch off history or training where the account allows, and ask the provider through its published route about deletion.

### What is the Paste Plan?

The Paste Plan is Dan Fitzpatrick's five-step response to customer data pasted into an AI tool: Stop the pasting, Scope what went in, Pull it back where you can, Decide who must be told, and Fix the cause. It is a suggested method, not a legal standard.

### Should you discipline an employee who pastes customer data into ChatGPT?

Usually not for a first incident. Thank them for reporting it, ask what they pasted, which account they used and what they were trying to do, then fix the gap. Reserve discipline for clear bad faith or breaking a rule you had written down and explained.

---
Source: [What to Do If an Employee Pastes Customer Data Into ChatGPT](https://business.theaieducator.io/posts/employee-pasted-customer-data-into-chatgpt)
Publisher: [The AI Educator](https://theaieducator.io)
