A data processing agreement for ChatGPT and other AI chat tools
· By Bee Flow · 4 min read
If an AI chat processes personal data for your company, you need a data processing agreement. When that applies, where to find it at OpenAI, Microsoft, Google and Anthropic, and what to check in it.

The accountant asks the office manager for a list of all data processing agreements. The bookkeeping software, the mail and the payroll are in the folder. Then she thinks of the three colleagues who use ChatGPT every day, on a subscription that was once paid with the company credit card. There is nothing on file for that one.
The short answer: if an AI service processes personal data on behalf of your company, you need a data processing agreement. That comes from Article 28 of the GDPR (in Dutch the AVG, and the agreement is a verwerkersovereenkomst). OpenAI offers one for ChatGPT Business, ChatGPT Enterprise and the API. The personal subscriptions are not on that list.
This article is not legal advice, and your own situation may differ. We checked the vendor terms below on their own pages on 14 November 2026. They change, so take a look yourself before you sign.
When you need one
The GDPR has two roles. The controller decides why and how personal data is processed. The processor does the work on the controller's behalf. If your team uses an AI chat for work that involves customer or staff data, you are usually the controller and the chat provider is your processor.
According to the Autoriteit Persoonsgegevens, the Dutch data protection authority, you and the processor then have to put the arrangements in writing, and you are both liable if the agreement is missing. A verbal arrangement does not count. And you stay responsible yourself, even when the processor drafted the text.
If no personal data ever passes through the chat, the provider processes no personal data for you either. In practice that rarely holds. One name in a mail is enough.
Where to find it for the common tools
At OpenAI you request the agreement through the form on the Enterprise privacy page. The text itself is the Data Processing Addendum. For customers in the European Economic Area the contracting party is OpenAI Ireland Ltd. The other companies OpenAI uses are in its sub-processor list.
At Microsoft, Copilot Chat falls under what Microsoft calls "enterprise data protection" as long as you sign in with a work or school account. Microsoft then acts as a processor under its Data Protection Addendum and the Product Terms, and does not use your prompts and responses to train the underlying models. That is how Microsoft describes it on its page about privacy in Copilot Chat. What the Dutch regulator and SURF published about Microsoft 365 Copilot in 2026 is in is Microsoft 365 Copilot GDPR compliant?.
At Google, prompts to Gemini in Workspace count as customer data under the Cloud Data Processing Addendum, and Google says they are not used to train models without permission. See the Generative AI in Google Workspace Privacy Hub. Using the Gemini app with a personal Google account falls under different terms.
At Anthropic, the data processing agreement, with standard contractual clauses, is automatically part of the Commercial Terms. Those cover Claude's Team and Enterprise plans and the API. The personal plans are outside them.
What to look for
Article 28(3) lists what an agreement has to settle. In its own investigation (in Dutch) the AP found agreements that did little more than repeat the text of the law. So check that it says, in concrete terms:
- that the provider processes your data only on your written instructions, and not for its own purposes
- that the people with access are bound to confidentiality
- which security measures the provider takes
- which sub-processors there are, and how and when you hear about a new one
- how the provider helps you with a request from a data subject and with a breach
- what happens to the data when the contract ends, and how you can check that it all holds
With AI chat, also look at training. Read what the agreement and the settings say about using your conversations to improve models, and check that it is off by default.
And check where the data goes. If the provider or its sub-processors are outside the European Economic Area, the transfer needs a legal basis, such as the European Commission's standard contractual clauses. For the United States there is also the EU-US Data Privacy Framework. The EU General Court upheld it in September 2025, but an appeal to the Court of Justice is still pending. Nobody knows yet how that ends.
Fewer parties between you and your data
With Bee Flow you can run the chat on your own server. The chat software itself then adds no extra processor. If you rent that server from a host, the host is usually a processor. For the language model you need an agreement with the provider you choose, or with nobody if the model runs on your own hardware.
The Compliance Center in Bee Flow keeps a record of processing activities that is built from your own configuration. External providers that Bee Flow sends data to are listed in it as processors, and per provider you record whether there is a data processing agreement and standard contractual clauses. You can export the record as a PDF. The Compliance Center is not in every licence; see the plans page.
More about the chat is on private AI chat, and about security on privacy and security, with the GDPR side in plain words on security, privacy and GDPR. Whether ChatGPT is safe at work is covered in is ChatGPT safe to use at work, when you also need an impact assessment in a DPIA for AI chat, and questions can go to contact.