Is AI GDPR compliant?

23 September 2026

No AI tool is GDPR compliant on its own, and none is automatically non-compliant either. Whether you can use ChatGPT, Claude, Microsoft Copilot or Google Gemini with your customers' details depends on two things: which version of the tool you use, and what your business does around it. For a UK small business the short answer is yes, you can use AI with customer data, as long as it's a business version and you do a handful of things the law already asks of you.

We set up AI and automation for small businesses. We're not data protection advisers, so this article explains what the law and the regulator's own guidance say rather than giving advice on your situation. Everything below links to the legislation or to the Information Commissioner's Office (ICO), the UK's data protection regulator, and for anything specific to your business the ICO's advice for small organisations is the place to start.

Is it legal to use AI with customer data in the UK?

Yes. There's no UK law that bans it, and there's no UK law written specifically for businesses using AI tools. In July 2026 the government told Parliament it prefers "context-specific regulation", looking at how AI is used sector by sector, and that it will legislate where it sees a need.

So the rules are the ones you already follow: the UK's version of GDPR, the Data Protection Act 2018 and, for marketing emails and cookies, the Privacy and Electronic Communications Regulations. The Data (Use and Access) Act 2025 changed those laws but didn't replace them. Most of its data protection changes came into force on 5 February 2026, and the ICO says all of them were in force by 19 June 2026. We've set out what the Data (Use and Access) Act changed for a small business in a separate article.

That's why "is ChatGPT GDPR compliant?" doesn't have a yes or no answer. GDPR is about what your business does with people's information. The tool can make it easy to do that properly, or make it impossible.

Why personal or business plan is the question that matters

Two ideas from the law explain almost everything here.

Your business is the controller of your customers' details. That just means you're the one responsible for them, in the same way you'd be responsible for a customer's paper file.

A company that handles those details on your behalf, following your instructions, is a processor. Think of a bookkeeper who works on your records but doesn't use them for anything of their own. The law says that when a processor handles personal data for you, there must be a contract saying it will only use the data on your instructions. For AI tools, that contract is usually called a data processing agreement.

Here's where AI plans split. The ICO's view is that if an AI company uses data "for their own purposes, they are a controller for that processing", and it has specifically rejected the argument that an AI company using a customer's data to improve its own models is only acting for that customer (ICO). On personal plans, the AI company runs your chats under its own privacy policy and may use them to improve its models. There's no contract saying it follows your instructions.

The line isn't free versus paid. ChatGPT Plus, ChatGPT Pro, Claude Pro and Claude Max all cost money, but they're personal plans (OpenAI, Anthropic). The versions that come with a data processing agreement are the business ones:

ChatGPT: OpenAI offers a data processing agreement for ChatGPT Business, ChatGPT Enterprise and its API, and doesn't train on them by default.
Claude: Anthropic's data processing agreement comes automatically with its business terms, which cover Claude Team, Claude Enterprise and the API, and those terms say it may not train models on your content.
Copilot: Microsoft Copilot used through a work account is covered by Microsoft's data protection terms, with Microsoft acting as your processor. Web searches Copilot runs for you are handled separately, with Microsoft responsible for them itself. The personal Copilot app on a home Microsoft account comes under Microsoft's consumer terms instead.
Gemini: Gemini inside a paid Google Workspace account is covered by Google's data processing terms and isn't used to train models without your permission. The personal Gemini app is different: Google's own privacy page for it says "Please don't enter confidential information" that you wouldn't want a reviewer to see.

A data processing agreement is the minimum, not the whole answer. The ICO says a contract doesn't on its own decide who is responsible for data - what actually happens to it does. It has also said that the companies behind closed AI models are unlikely to be acting purely for the businesses that use them, so it sees responsibility as likely to be shared. None of that makes a business plan the wrong choice. It's still the version built to be used with other people's details, and the personal plans aren't.

We've written separately about the training settings on each plan in whether AI trains on your business data.

If you'd like everyone in the business moved onto the right kind of plan, with the settings checked, we're happy to set that up for you.

Can I put customer data into ChatGPT?

On a business plan, yes, as long as you've done the other things in this article. On a personal plan, it's hard to see how it could meet the rules. The AI company uses those chats for its own purposes, so you'd be handing your customers' details to another business, with no contract limiting what it does with them.

The more common problem is staff doing it without anyone knowing. People paste a customer's email into their own free account to get a better reply written, with the best of intentions. The law says your business must take steps to make sure anyone handling personal data for you only does so on your instructions. In practice that means deciding which accounts work happens in and writing it down, which is simpler than trying to ban AI altogether.

If customer details have already gone into a personal account, that may count as a personal data breach. The law says every breach must be recorded, and reported to the ICO without undue delay - within 72 hours of finding out, where that's feasible - unless it's unlikely to put anyone at risk. If the risk to people is high, you also have to tell the people affected. Whether a particular case needs reporting depends on what was shared, and the ICO's guide to personal data breaches walks through how to judge it.

Do I need customers' consent to use AI?

Usually not. Every use of personal data needs a lawful reason, and consent is only one of the reasons the law lists. Most everyday business use rests on something else, such as needing the details to do what the customer asked, or what the law calls legitimate interests. Sensitive details, such as health information, need more, and we come to those further down. Recording calls is a case of its own, which we cover in whether you need consent to use an AI notetaker.

Our reading is that using a business AI tool to help draft a reply to a customer who wrote to you is a new way of doing the same job, not a new use of their details. The ICO hasn't published an example on this, so treat that as our view rather than the regulator's.

Using details you already hold for something new is different. If you wanted AI to go through years of customer emails to score people or find sales leads, the new purpose has to fit with the reason you collected the data, or you need people's consent to it, and the ICO says you must tell people before you start. If you rely on legitimate interests, the ICO says you must pass a three-part test before you start, and that you should write it down. The law doesn't require the writing itself, but you do have to be able to show you did the test.

One thing to be wary of: the 2025 Act added something called "recognised legitimate interests". It sounds as if it might help, but it only covers five public-interest purposes, such as preventing crime, emergencies and safeguarding vulnerable people. Everyday business use of AI isn't one of them.

What does my privacy notice need to say?

Your privacy notice is the page that tells people what you do with their details. The ICO says it must list who you share personal data with, including companies that handle it for you, either by name or by type. The law itself also requires you to say whether it goes outside the UK. So once an AI company is processing customer details for you, it belongs in the notice.

We haven't found anything in the law or the ICO's guidance that says you must label an email as written with AI's help when a person has checked it and sent it. The duty is to be open about what happens to people's details, not to stamp every message.

Does it matter that the AI company is in the US?

It matters, but it doesn't stop you. The law restricts sending personal data to organisations outside the UK unless there's an approved route.

The main route to the US is the UK-US data bridge, in place since 12 October 2023. It only covers US companies that are actively signed up to it. When we checked the official list on 23 September 2026, Microsoft and Google were on it and OpenAI and Anthropic weren't.

That doesn't make OpenAI or Anthropic off limits. Their business data processing agreements use a different approved route instead: standard contract terms for international transfers, with the UK's own addendum (OpenAI, Anthropic). When you rely on those, the law expects you, acting reasonably and proportionately, to be satisfied that the protection people get would not be materially lower than in the UK.

Some business plans also store data in the UK. Microsoft commits to keeping Copilot chats in the UK for business accounts whose data location is the UK, under its standard terms. OpenAI offers UK storage on its Enterprise plan for new customers, but it isn't the default elsewhere. It's worth knowing about, but where the servers sit isn't what makes a transfer lawful: the rules are about which organisation receives the data.

What about AI that makes decisions on its own?

Most small business uses of AI don't come near this. Sorting enquiries into the right folder, or drafting a reply that a person reads before sending, isn't the kind of decision the rules below are about.

The rules apply when a decision with a legal or similarly serious effect on someone is made with no meaningful human involvement, such as automatically rejecting a job applicant or turning down credit. Since the 2025 Act, that's generally allowed, but only with safeguards, including a way for the person to get a human to look again and to challenge the decision. It's still tightly restricted where the decision uses sensitive details such as health information, and it can't rest on the new "recognised legitimate interests". The ICO's updated guidance on this is still a draft, with the final version listed as due in Winter 2026. There's more on where the line falls between a tool and something that acts by itself in what an AI agent is.

Do I need a data protection impact assessment?

A data protection impact assessment, or DPIA, is a written check of the risks before you start something. The law requires one before processing that's likely to be high risk.

The ICO's pages point in slightly different directions here. Its official list says AI counts as "innovative technology", and that makes a DPIA compulsory when combined with another risk factor, such as sensitive information, vulnerable people, scoring people or large volumes. Its older guidance on AI goes further, saying most uses of AI will be high risk, and that if you decide you don't need a DPIA you still need to write down how you decided.

The reading that satisfies both is to do a short written check for each way you use AI with personal data, and a full assessment where any of those risk factors apply. Our article on when you need a DPIA to use AI goes through the ICO's screening test and the uses that clearly need one.

The things the law already asked of you

Using AI doesn't create these duties, but it's a good moment to check them, because each one now includes your AI tools:

A written record of what you do with personal data. Businesses under 250 staff are excused only for occasional, low-risk processing that doesn't involve sensitive details, so regular use such as handling customer emails still needs recording.
The ICO fee. A business with 10 or fewer staff, or a turnover of £632,000 or less, is in the lowest band, £52 a year, £5 less by direct debit, unless everything it does is exempt.
A way to complain. Since 19 June 2026 you must make it easy for people to complain about how you use their data, acknowledge a complaint within 30 days and respond without undue delay.
Extra care with sensitive details. Health, ethnicity, religion and similar information need an extra legal condition on top of the usual reason. The ICO points out that even a broken leg counts as health information.
People's rights still reach AI. The ICO says that what an AI tool produces about a person, once you store it, is generally covered by people's rights, such as the right to see it.

The regulator's guidance is catching up

Part of the reason this is confusing is that the official guidance is mid-change. The ICO's main guidance on AI was last updated in March 2023 and carries a notice saying it's under review because of the 2025 Act. The ICO is now required by law to write a code of practice on AI, and no draft had been published when we checked. And since 30 September 2026 the ICO's work has passed to a new body, the Information Commission, which is still known as the ICO.

None of that changes the law described above. It does mean many pages you'll find online were written before the 2025 Act, so check the date on anything you rely on, including this one.

What should you do next?

If you're already using AI with customer details, the first step is simply finding out which accounts it's happening in. If it's all on business plans, most of what's left is paperwork you may already have: a line in your privacy notice, a note in your records and a short written check.

If it's happening on personal accounts, that's the part worth fixing first, and it's mostly a setup job rather than a legal one. We're happy to move your team onto business accounts with training switched off, and to build any automations so they only send the AI the details each task actually needs, with old chats and records cleared out on a schedule you choose. You can see what we set up for small businesses, or book a free call and we'll look at where AI already touches your customer data.

Sources

Checked on 23 September 2026. The move to the Information Commission was checked on 1 October 2026.

The next step

The cost of waiting is not zero.

Somewhere in your market, a business the same size as yours has stopped writing follow-ups by hand. They are quoting faster than you, chasing money you are still chasing manually, and starting Monday already knowing where they stand. That gap does not stay the same size.

Half an hour on the phone. You describe your week, we tell you honestly which parts of it are worth automating and which are cheaper left alone. No pitch, and nothing to sign - you decide whether you want to take it further.

Book a free call

Or just ring us on 07816 970 768, or email hello@seamwork.co.uk.