Is ChatGPT safe to use at work? What happens to what you type
· By Bee Flow · 6 min read
Whether ChatGPT is safe at work depends mostly on the account and on what you agree with your team. What happens to your text, which settings and subscriptions change that, and what the Dutch regulator advises.

On Monday morning someone from accounts asks whether she can use ChatGPT to write a payment reminder to a customer. She has been doing it at home for months, on her own free account. She wants to know whether that is allowed at work, and whether it is safe.
The short answer: it mostly depends on the account you use and on what you agree with your team. With an employee's personal account, your company has no agreement with OpenAI, and conversations may be used to train models by default. With a business subscription, training is off by default and you can sign a data processing agreement. Either way, what you type ends up on OpenAI's servers.
This article is not legal advice, and your own situation may differ. We checked everything below on 14 November 2026 against OpenAI's own pages and those of the Autoriteit Persoonsgegevens, the Dutch data protection authority.
What happens to what you type
A question to ChatGPT goes to OpenAI, is processed there and stays in your history. When you delete a conversation it disappears from your list at once. According to OpenAI it is then permanently deleted within 30 days, unless it had already been disconnected from your account or has to be kept for security or legal reasons.
On personal accounts, OpenAI may use your conversations to improve its models by default. You switch that off in the settings under Data controls, with the toggle "Improve the model for everyone". It then applies to new conversations. If you give an answer a thumbs up or down, the same page says the whole conversation may still be used for training.
There is also a temporary chat. It is not used to train models, and OpenAI keeps a copy for at most 30 days for safety checks. It is still a personal account with no agreement with your company.
What a business subscription changes
For ChatGPT Business (called ChatGPT Team until August 2025), ChatGPT Enterprise and the API, OpenAI says it does not use inputs and outputs for training by default. For the same products OpenAI signs a data processing agreement, and for customers in the European Economic Area the contracting party is OpenAI Ireland Ltd.
A business account does not make the use compliant by itself. You get a contract and a fixed default for training, and you build the rest on top of that.
What the regulator says
The Autoriteit Persoonsgegevens (AP) warned in August 2024 that it was receiving data breach reports because employees had put personal data into AI chatbots. At a GP practice it was patients' medical data, at a telecom company a file with customer addresses. Often the staff used the chatbot on their own initiative and against what they had agreed with their employer.
In the AP's view such input can be a data breach. Reporting it to the AP and to the people concerned is then mandatory in many cases. Its advice to organisations is to agree clearly which data may and may not go into a chatbot, and possibly to arrange with the provider that the data entered is not stored.
What a company should arrange
Start with the choice of account. If staff use their own free accounts, you have no contract and no view of what happens. A business subscription, from OpenAI or another vendor, gives you an agreement. What belongs in it is in a data processing agreement for ChatGPT and other AI chat tools.
Then write down what is and is not allowed. A few rules are enough, as long as they are concrete. A made-up example: rewriting and summarising text is fine, pasting customer mails with names, citizen service numbers (BSN, the Dutch personal ID number) or health data is not. Add what someone should do when it happens anyway. Under Article 33 of the GDPR a breach goes to the supervisory authority without undue delay and where feasible within 72 hours, so someone needs to hear about it quickly.
Add the use to your record of processing activities, and check whether you need a data protection impact assessment (DPIA). That is the subject of a DPIA for AI chat in a small company.
Also expect the rules to be forgotten now and then. On a busy day everyone pastes something that should not be there.
Is ChatGPT Business GDPR compliant?
No product is GDPR compliant by itself, and no vendor can make your use compliant for you. ChatGPT Business gives you two things a personal account lacks: a data processing agreement with OpenAI Ireland Ltd. and training that is off by default. The rest stays with you as the controller:
- a purpose and a legal basis for the personal data that goes in,
- a check of where the data is processed and which subprocessors are involved,
- an entry in your record of processing activities, and a DPIA when the use carries a high risk,
- rules for staff about what may go into a prompt.
The same holds for Microsoft's assistant. Is Microsoft 365 Copilot GDPR compliant? sums up what the Dutch regulator and SURF published about it in 2026.
Are you allowed to use ChatGPT for work?
No law forbids it. In practice three things decide the answer.
What your employer has agreed. If there is a rule, follow it. If there is none, ask. A missing rule is not permission to paste a client file.
Which data it is. Rewriting a text without names or confidential details is rarely a problem. Client names, BSNs, health data and payroll figures only belong in a business account that is covered by an agreement, and only where your company's rules allow it.
The rules of your profession. Professional bodies add their own conditions. For accountants, the guide on applied AI that the NBA and NOREA published in June 2026 asks that client data stays in a secured, closed environment. What that means for a firm is on AI for accounting and tax firms.
If you want to keep the chat in your own hands
Bee Flow is an AI workspace you can run on your own server. Conversations are then stored in your own database, and your organisation decides how long they are kept. You choose which language model answers: one on your own hardware, an open model at a European host or a large cloud model. With a cloud model the text still goes to that provider, and you need a data processing agreement with them.
Chat messages first pass a filter on your own server. It recognises names, IBANs, BSNs and other personal data and holds the message back before it reaches the model. It needs the guard service, which the default install includes. Detecting and blocking is in every licence. On the Pro licence the data can be replaced before the model sees it and put back in the answer. Detection is machine learning and can miss. How it works and where it falls short is in what happens when staff paste customer data into an AI chat.
More about the chat is on private AI chat, and about security on privacy and security. How Bee Flow compares with OpenAI's own business plan is on the ChatGPT Business alternative page. If you are comparing vendors, choosing a GDPR-proof alternative to ChatGPT can help. Questions can go to contact.
More on this question
- What happens when staff paste customer data into an AI chat
- A data processing agreement for ChatGPT and other AI chat tools
- A DPIA for AI chat in a small company
- Choosing a GDPR-proof alternative to ChatGPT: five questions for the vendor
- Is Microsoft 365 Copilot GDPR compliant? What the regulators have said