AI tools are not GDPR compliant or non-compliant on their own. How your business sets them up and uses them is what counts. Here is a practical checklist for keeping client data safe.
"Is ChatGPT GDPR compliant?" is one of the most searched AI questions among UK businesses. The honest answer is that no AI tool is compliant or non-compliant by itself. What matters is which version you use, how it is configured and what your team puts into it.
This is Step 5 of our six-step guide to AI implementation. It is general guidance, not legal advice, so take specialist advice for your own situation.
Who is responsible?
Under UK GDPR, when your team puts personal data into an AI tool for business purposes, your business is normally the controller of that data. The AI provider usually acts as a processor on your behalf.
That means the responsibility for using the data lawfully, securely and transparently stays with you, whichever tool you use. The Information Commissioner's Office (ICO) publishes guidance on AI and data protection that is worth reading alongside this article.
Business accounts versus personal accounts
This is the single most important decision.
Business and enterprise plans from the major providers typically include a data processing agreement, admin controls, and terms stating that your business data is not used to train their models by default.
Free and personal accounts are designed for individuals. Terms around data use, retention and training differ, and you have no admin oversight of what staff put in.
If staff are using personal AI accounts for client work, that is usually the first thing to fix. Always check each provider's current terms, as they change.
A practical AI data protection checklist
- Use business accounts only for work involving personal or client data.
- Sign the data processing agreement your provider offers, and keep a copy.
- Check where data is processed and stored, and what safeguards cover any transfer outside the UK.
- Review retention and training settings. Confirm business data is not used for model training and understand how long it is kept.
- Set clear data rules for staff. Special category data, such as health information, needs particular care and often should not go into general AI tools at all.
- Limit what connected AI can reach. When AI is connected to SharePoint, Drive or your CRM, it should only access what each user is already allowed to see.
- Turn on strong sign-in. Single sign-on and multi-factor authentication for every AI account.
- Update your privacy notice if AI is used to process customer data in new ways.
- Carry out a DPIA for higher-risk uses. A data protection impact assessment is expected where processing is likely to be high risk, for example customer-facing AI or automated decisions about people.
- Train your team and include AI in your data breach procedures.
Security risks specific to AI
As AI connects to more systems, a few new risks appear:
- Over-sharing through permissions. AI connected to a badly permissioned drive can surface files people should not see.
- Prompt injection. Content from outside, such as an email or web page, can contain hidden instructions that try to redirect an AI tool. This matters most for AI agents that take actions.
- Shadow AI. Staff using unapproved tools because the approved ones are hard to access.
The fixes are the same principles as good IT security: least privilege, approved tools, human approval for important actions, and logging.
Compliance as a selling point
Many of your clients will soon ask how you use AI with their data, if they are not asking already. A clear policy, business-grade tools and a simple register of AI use give you a confident answer. For firms in regulated sectors, that can be a genuine advantage in tenders.
How we can help
We set up AI tools on business-grade accounts inside your own systems, with permissions, policies and documentation in place. Our AI Ops Audit includes a review of where AI currently touches personal data in your business.
Next in the series: AI agents for business: automating whole workflows.