The question has no yes-or-no answer, and that is not evasion. GDPR compliance is not a badge a product carries. It is a description of a particular use, by a particular organisation, for a particular purpose — which means the answer depends more on you than on OpenAI.
General information, not legal advice. If this decision matters to your organisation, it belongs with your DPO or counsel.
You are the controller
This is the sentence that reorganises the whole question. When you type a customer’s details into an assistant, you decide the purpose and the means of that processing, which makes you the controller. The AI company is your processor. The obligations — lawful basis, transparency, minimisation, erasure — are yours.
So “is ChatGPT GDPR compliant” is close to asking whether a filing cabinet is compliant. What matters is whether you had a reason to put that file in it, whether the person knows, and whether you can get it out again.
Asking whether the tool is compliant is like asking whether a filing cabinet is. It depends what you put in it.
What OpenAI provides
- •A data processing agreement. The Article 28 contract you need before a processor may handle personal data on your behalf. It has to be entered into — a personal account does not have one by default.
- •Transfer mechanisms. Standard Contractual Clauses for the transfer of data outside the EEA, which is the mechanism most transatlantic arrangements rest on.
- •Terms that exclude training on business and API data by default, and a setting to turn it off on consumer accounts. These differ by product, and the difference is the thing to check.
- •Deletion and export controls in the account, which help you answer a subject access or erasure request.
That is a serious set of pieces, and it is more than many vendors offer. It also does none of your work for you.
Where organisations actually fail
- •Personal accounts at work. Staff pasting customer emails into their own logins. No DPA, no register entry, no retention control. This is the most common failure by a wide margin and it is a policy problem, not a technical one.
- •No lawful basis for the new purpose. Data collected to deliver a service, then used to prompt a model, is a second purpose that needs its own justification.
- •No answer to erasure. Article 17 gives people a right to deletion. If you cannot say where a customer’s details went after they were pasted into a chat, you cannot honour it.
- •Silence in the privacy notice. If you send personal data to an AI processor and your notice does not say so, the transparency obligation is unmet whatever else you got right.
Regulators have been active here
European authorities have taken a close interest in generative AI since 2023 — Italy’s Garante temporarily suspended ChatGPT in the country and later issued a fine, and the European Data Protection Board convened a task force on the subject. The recurring themes are lawful basis for training data, transparency, and how erasure and accuracy rights apply to a model rather than to a database.
None of that makes the tool unusable. It does mean the area is unsettled, and that “the vendor says it is compliant” is not a position you would want to defend on its own.
The minimisation argument
Article 5 asks for data minimisation: adequate, relevant, and limited to what is necessary. Applied to an assistant, that principle points somewhere specific — most prompts do not need the personal data in them to get a useful answer. “Draft a reply to this complaint from Maria Schneider at 14 Kastanienallee” and the same request with a stand-in name and street produce the same letter.
This is the design Secure AI is built on. Names, addresses, phone numbers, email addresses, national identifiers, card and account numbers are detected and replaced with realistic stand-ins before the message reaches a model; the reply comes back with your real values restored. The processor answers a question about somebody who does not exist, which is the strongest form of minimisation available: not a promise about handling personal data well, but less personal data leaving at all.
It does not make you compliant on its own. You still need your basis, your notice and your register. It removes the failure mode that no contract can fix — data that has already gone somewhere you cannot reach.
Questions
Is the free ChatGPT account GDPR compliant for business use?
Treat it as not. A personal account generally has no data processing agreement behind it, which is the first requirement for using a processor at all. Business and enterprise arrangements exist for this reason.
Does turning off training make it compliant?
It helps with one issue and leaves the rest. You still need a lawful basis, a DPA, a transfer mechanism, transparency, and an answer for erasure. The toggle addresses none of those.
Does the EU AI Act replace any of this?
No — it sits alongside. The AI Act governs how AI systems may be built and placed on the market; GDPR governs personal data wherever it goes. A use can satisfy one and breach the other.
What is the single most useful thing to do this week?
Write down which assistants your staff are using and on whose accounts. Almost every organisation that does this finds at least one answer it did not expect, and you cannot govern a transfer you do not know is happening.
Try Secure AI free
Frontier AI chat — encrypted by default, with your identity stripped before any request reaches the provider.