Is ChatGPT safe for businesses?

Is ChatGPT safe for businesses?

ChatGPT can be safe for your business, if you arrange three things: the right account, agreements with the provider and agreements with your team. The tool itself is not the biggest risk. The risk lies in what people type into it, and under which conditions that happens.

This is not theory. The Dutch Data Protection Authority (Autoriteit Persoonsgegevens, AP) received reports of data breaches caused by employees entering personal data into an AI chatbot. In one case, an employee of a GP practice entered patients’ medical data into a chatbot. In another case, it involved a file with addresses of customers of a telecom company. According to the AP, this often happens on employees’ own initiative, against the agreements.

The good news: with a few choices you remove a lot of risk. Those choices are set out below.

What is the difference between a free and a business account?

The biggest difference is what happens to your conversations. With ChatGPT’s personal plans (free, Go, Plus and Pro), OpenAI may use your conversations to improve its models. You can turn that off in the settings, under Data controls, with the ‘Improve the model for everyone’ switch.

With the business plans, ChatGPT Business and ChatGPT Enterprise, OpenAI does not use your data to train models by default. You also get an admin environment. In it, an administrator decides who has access.

So a paid personal account, such as Plus, is not a business account. You do pay, but it remains an account for one person, under the consumer terms. What the business plans cost, you can read in ChatGPT, Copilot or Claude: which AI suits your business?

What does a data processing agreement arrange?

With a data processing agreement you record how another party handles the personal data you supply. The GDPR requires such an agreement when a party processes personal data on your behalf. Does your team put data about customers or colleagues into ChatGPT? Then OpenAI is such a party.

OpenAI offers a data processing agreement for ChatGPT Business, ChatGPT Enterprise and the API. OpenAI calls it a data processing addendum. You conclude it via a form from OpenAI. What belongs in such an agreement, you can read under data processing agreement.

Two things to know:

• You remain responsible. Even with a data processing agreement, you as a business are responsible for the personal data. The agreement does not make every use acceptable. You still need a good reason to use data, and you don’t use more data than necessary.
• Check where the data is stored. With some business plans you can choose storage in Europe. Ask what is possible with your plan.

Which data does not belong in ChatGPT?

Data you wouldn’t just email to an unknown company does not belong in it. Even with a business account, caution is sensible. Without good agreements, you don’t put this data into it:

• sensitive personal data, such as medical data or a citizen service number (burgerservicenummer, BSN);
• customer files with names, addresses or phone numbers;
• HR matters, such as appraisals, sickness absence or salaries;
• passwords, login codes and API keys;
• confidential information from others, such as a contract in which you agreed to confidentiality.

Often you don’t need that data anyway. Do you want help with a difficult email to a customer? Then first remove the name and other details. If you replace them with something like ‘customer A’, that is called pseudonymisation. Note: if the person remains identifiable from the rest of the text, it is still personal data.

What happens if an employee enters personal data anyway?

Then it may be a data breach. According to the AP, there is a data breach when an employee enters personal data into an AI chatbot, against the employer’s agreements. The company behind the chatbot then receives data it was not allowed to have access to.

In many cases you must report a data breach to the AP, no later than 72 hours after you discover it. Sometimes you also have to inform the people whose data it is. Record every data breach in your own overview of data breaches, even if you don’t have to report it.

And what if the business itself allows the use? Then, according to the AP, it is not a data breach. But it is often still not permitted by law. That is why you need both: a business account with a data processing agreement, and agreements about which data may and may not go into it.

May an employee use ChatGPT for work?

Yes, if you as the employer allow it and make agreements about it. The AP advises businesses to make clear agreements. Is it allowed? And which data may and may not go into it? Also agree which account your team works with.

Without agreements, everyone chooses their own tool and their own account. Then you don’t know which AI tools are used in your business, and under which conditions. Does that sound familiar? Then read Your team uses ChatGPT without agreements.

Good agreements fit on one page. Think of:

• which AI tools are allowed, and with which account;
• which data does not belong in them;
• that a person checks every result before it goes out the door;
• who you can turn to with a question or a mistake.

An example you can adopt is in AI policy on one page: an example for SMEs.

There is another reason to arrange this. Does your team use AI for work? Then under the EU AI Act you must take measures for your people’s AI literacy. Explanation and agreements are part of that. Read more in AI literacy is mandatory.

Does this also apply to Copilot and Claude?

Yes. For every AI assistant you ask the same three questions: which account, which agreements with the provider and what goes into it. The difference is in the details.

Microsoft Copilot

If you sign in with your company’s work account, Copilot Chat falls under Microsoft’s business terms, with a data processing agreement (the Data Protection Addendum). Microsoft then does not use your prompts and answers to train the underlying models. In Copilot Chat you see a green shield when that protection is on. If Copilot searches the internet, those search queries go to Bing. Different terms apply to that.

Claude

With Anthropic’s business plans, Claude Team and Claude Enterprise, Anthropic does not train on your data by default. The data processing agreement is part of the business terms. With the personal plans (Free, Pro and Max), you choose yourself whether your conversations may be used to improve models. If you say yes, Anthropic keeps those conversations for up to five years.

Which of the three suits you best depends mainly on the programs you already work in. You can read about that in ChatGPT, Copilot or Claude: which AI suits your business?

How do you make sure your team uses ChatGPT safely?

With a few fixed steps. You don’t have to take them all at once.

1. Choose one business account for the whole team, instead of separate personal accounts.
2. Conclude the data processing agreement and keep it with your other privacy documents.
3. Secure access with two-step verification. Remove the account when someone leaves the company.
4. Make agreements on one page and discuss them with the team.
5. Explain where AI goes wrong. ChatGPT can give answers that sound right but aren’t. That is called a hallucination. So have a person check every result.
6. Keep an overview of the AI tools you use, what for and who the owner is. That is called an AI register.

How I help you with this

I give the AI policy & training session at your business. In half a day I explain to your team what AI can do, where it goes wrong and how to use it safely. Together we create clear rules for your business, and an overview of the AI tools your team uses. Does your team then want to practise on their own work for a full day? Then there is the one-day team training. We discuss what it costs in the first conversation, so you know in advance where you stand. Read more on AI training.

Do you want AI built into the systems you already use? Then read about AI consultancy. I don’t sell licences and I receive no commission from OpenAI, Microsoft or Anthropic. Not sure what suits your business? Schedule a conversation.

Frequently asked questions

Sources: AP: Beware, using an AI chatbot can lead to data breaches (in Dutch) (6 August 2024), AP: Data processing agreement (in Dutch), AP: How to report a data breach (in Dutch), AP: Data breach, what you must do (in Dutch),OpenAI: How your data is used to improve model performance, OpenAI: Enterprise privacy, OpenAI: plans, Microsoft: Copilot Chat privacy and protections, Anthropic: Claude plans, Anthropic: Updates to Consumer Terms and Anthropic: Data Processing Addendum. Consulted on 7 October 2026.