"We probably cannot use AI, surely it is not approved under GDPR?" It is a common reaction, and it rests on a misunderstanding. GDPR is not a list of approved and banned technologies. It is a set of rules for how personal data is processed. This article goes through six misunderstandings that often come up at Swedish companies, and what applies in practice. The text is general guidance and not legal advice. What applies to your business should be assessed with qualified advice, for example from your data protection officer or a lawyer.
Misunderstanding 1: "AI and GDPR do not go together"
GDPR regulates how personal data is processed, whether that happens in a spreadsheet, a ticketing system or an AI agent. The same basic questions apply: why do you process the data, what legal basis do you have, how much do you need and for how long, who has access and what do the agreements say? An AI agent is neither exempt nor specially banned. It is one more system where personal data can appear, so it should be mapped in the same way as the others.
Misunderstanding 2: "The provider said they follow GDPR, so we are done"
A statement on a website is not an agreement. If the provider processes personal data on your behalf, it is normally your data processor, and a written agreement is required. Ask for it before you decide and read what it says about:
- which data is processed and for what purpose
- which sub-processors are used
- how and when data is deleted
- what happens in the event of a personal data incident
- what applies to transfers outside the EEA
Responsibility for the processing being lawful still rests with you as the controller. More questions to ask are in the vendor selection checklist.
Misunderstanding 3: "Where data is located does not matter if the provider is serious"
Location matters, but it is only part of the picture. If data is processed or made accessible outside the EEA, you need to assess the rules for transfers to third countries. That also applies to sub-processors, support access and tools the provider uses. So do not ask only where storage happens, but where the data is processed, who can reach it and on what basis. Ask for the answer in writing and have an adviser interpret it.
Misunderstanding 4: "We do not need to say it is an AI"
Openness is good practice, and when personal data is collected there are requirements to inform people about the processing. The customer should understand who or what they are talking to, which information is saved and how to reach a person. It is also good for trust. A clearly labelled AI agent that offers a route to staff creates fewer surprises than one that tries to sound like a person. You can read about how and when the handover happens in when AI should escalate to a human.
Misunderstanding 5: "More data is better for the agent"
Data minimisation is a basic principle: do not collect more than you need for the purpose. An agent that answers questions about opening hours and prices needs no personal identity numbers, and a booking dialogue needs no more details than the booking requires. Decide which fields the agent may ask for, and instruct it not to request sensitive information unnecessarily. That reduces both the risk and the amount of data you have to manage. See how knowledge is configured for what the agent answers from.
Misunderstanding 6: "Once it is set up, we are done"
After setup, the ongoing work remains. Conversations are saved, customers can request access or deletion, providers change sub-processors and your own business changes. Decide who is responsible for answering requests from data subjects, how often you review what is stored and for how long, and how you test that deletion actually works in each system. GDPR and data retention: practical rules goes through that part in more detail.
How to move forward
- Map the flow. Which data comes in, where does it end up and which providers are involved?
- Set purpose and limits. What may the agent ask for, and what should it never request?
- Request agreements and terms. A data processing agreement, sub-processors and deletion routines.
- Inform clearly. Let the visitor know it is an AI agent and how to reach a person.
- Test before launch. See how to test an AI agent and include privacy questions.
- Get an assessment. Let qualified advice decide what applies to you.
Practical starting points are in the guides, in the frequently asked questions and in the glossary for terms such as data processor. If you have questions about how Aveexia works, you can contact us.
