The short answer

Can I put customer data into an AI chatbot?

Customer data may only go into an AI chatbot if your business has arranged it in advance: a business plan, a data processing agreement with the provider and agreements on which data may go in. If an employee puts it in anyway, against those agreements, the Dutch Data Protection Authority calls that a data breach.

Customer data is almost always personal data: a name, an address, an email address, a complaint with a name attached. The GDPR applies to it, also when you type it into a chat window. Whoever puts it into a chatbot hands it to another company.

Business data with no person attached, such as a price list or a quote text without a name, is not personal data. There the question is a different one: do you want that data sitting on another company’s servers?

The occasion

What goes wrong when an employee puts customer data into a chatbot?

When an employee puts customer data into a chatbot, your business loses control over that data. Most companies behind chatbots store everything you enter, writes the Dutch Data Protection Authority (AP).

In August 2024 the AP issued a warning after reports from two organisations. At a GP practice, an employee had entered patients’ medical data into a chatbot, against the agreements. At a telecom company it was a file containing, among other things, customers’ addresses.

According to the NOS, that data ended up on the servers of American tech companies. There it can be used to train the model.

It did not stop at those two. According to Accountant.nl, the AP received dozens of reports in 2024 and 2025 of sensitive information shared through chatbots. In 2025 there were more than the year before.

The rules

When is it a data breach?

It is a data breach when personal data ends up somewhere it was not allowed or not meant to go. An employee who types customer data into a chatbot against the agreements therefore causes a data breach, the AP writes.

If your business actually prescribed the use, it is not a data breach. According to the AP it is then often still against the law, if nothing else has been arranged.

With a data breach, according to the AP, you sometimes have to report it to the AP within 72 hours, and sometimes also inform the people it concerns. Every data breach, even a small one, belongs in your own data breach register.

What the GDPR asks

What do you need to arrange before it is allowed?

Before customer data may go into a chatbot, you arrange the contract with the provider and the agreements with your own people. That comes down to four things.

  • A business plan, no personal accounts: an account in the company’s name, with a contract. In it, according to the AP, you can agree that the provider does not store your input to train its model.
  • A data processing agreement: if you have another company process personal data for you, the AP says you must sign a data processing agreement with that company. That also applies to the provider of your chatbot.
  • Agreements on what may go in: the AP advises setting down which data may and may not be entered into AI systems.
  • What you do after a slip: who hears about it, who decides on reporting and who records it in the register.

Even when all of this is in order, it remains your responsibility. According to the AP, as the controller you always remain responsible for the personal data you process, also when you outsource the work.

My reading

Who carries the responsibility when it goes wrong?

The responsibility lies with the business that holds its customer’s data, never with the chatbot provider. That is what the law says, and it is how your customer sees it too.

In my book De Informatiemotor I write about the rented engine. Almost everyone works on AI from another company, and the most sensitive things you have flow across it. On how the providers handle that, I write: “Usually it is reasonably well arranged, but usually is not a policy.”

And do not forget what already applied before the engine arrived: personal data falls under the GDPR, and professional secrecy remains professional secrecy, also in a prompt. Your customer trusts you, not your landlord.Paul Hoffman

A subscription is therefore not yet a policy. On the ladder in the book you can hear from the language where a business stands: “Whoever says ‘we have a subscription to it’ is on rung two, however proud it sounds.” Rung three only begins where there is an owner and a policy.

In practice

What do you arrange this week?

This week you write one page of agreements on AI and customer data. It says which chatbot your business uses, whose name the account is in and which data never goes in.

Also put on it what someone does if it happens anyway: report it the same day to one fixed person, who decides whether it goes to the AP. Make it easy to report a slip, so you hear about it in time.

A business licence, a data processing agreement and one page of agreements belong to the first level of AI in a business. Read in the article on the five levels of AI what else belongs to that level, and take the short check there.

Answered briefly

Frequently asked questions

Can I have a customer email rewritten if I remove the name?

Yes, if nothing is left in it by which the customer can be recognised. Removing the name is often not enough: an address, an order number or a description of what happened also gives away who it is. If in doubt, treat it as customer data and use only the business plan you have agreements for.

Is a paid subscription enough?

No, not by itself. A paid account in the name of one employee is still that person’s own account, with no agreements between your business and the provider. What counts is a business plan your company signs the contract for, with a data processing agreement. Also check in the terms whether the provider uses your input to train its model.

What do I do if an employee has already done it?

Treat it as a possible data breach. Write down what was entered, when and in which account, and record it in your data breach register. Then check whether you must report it: according to the AP that sometimes has to happen within 72 hours, and sometimes you must also inform the people concerned. Ask the provider whether the data can be removed from the account.

Does this also apply to data about my own staff?

Yes. Sick leave notices, reviews and contracts of your employees are personal data just as much, and often more sensitive than customer data. So your agreements should cover not only customer data but all data about people. Data about someone’s health falls under even stricter rules.

Do I need a lawyer for this?

Not to get started. A business plan, the provider’s data processing agreement and one page of agreements you can arrange yourself. Do bring in a lawyer if you process medical data, if you are bound by professional secrecy or if you are unsure whether you may use the data for this. This article is not legal advice.

Editorial note

Sources

The idea, the concept and the corrections of substance are Paul Hoffman’s. AI corrected the language, ordered the text and improved readability. The quotes come from his book De Informatiemotor (2026), translated. The facts were checked on November 10, 2026 against the sources below. This article is not legal advice.

  1. Autoriteit Persoonsgegevens (Dutch Data Protection Authority), August 2024: Caution: use of AI chatbot may lead to data breaches
  2. Autoriteit Persoonsgegevens (Dutch Data Protection Authority), accessed November 10, 2026: Controller and processor (in Dutch)
  3. Autoriteit Persoonsgegevens (Dutch Data Protection Authority), accessed November 10, 2026: Data breach? This is what you have to do
  4. NOS, August 7, 2024: Privacy watchdog: be careful using AI chatbots, it went wrong at a GP practice (in Dutch)
  5. Accountant.nl, December 30, 2025: AP raises the alarm over data breaches through AI (in Dutch)
NEXT STEP

Start with a conversation.

You work with me. So a first conversation is with me as well. An hour, no obligation, at a moment that suits you.