Is it legal to use AI for customer data? A GDPR guide for Danish businesses
AI can be used legally for customer data after a specific assessment. Get an overview of purposes, data types, roles, legal bases and data transfers.
Table of contents
Is it legal to use AI for customer data under GDPR?
AI can be used legally for customer data, but it requires a specific assessment of the purpose, the data types, the roles of the parties and the risks of the processing. When the data is personal data, GDPR applies; a data processing agreement is necessary when the vendor processes information on your behalf.
Clarify the legal basis, necessity, information to customers, security, storage and any transfers. Special data types and uses can trigger further requirements. A short checklist or a signed agreement is therefore not in itself a guarantee of legality.
Why the question is coming up right now
A great many Danish businesses have, over the past year, put AI tools into their daily workflows: customer service chatbots, AI assistants that read inboxes, agents that enrich CRM data or write draft replies. It often happens quickly, because the tools are easy to adopt: an employee signs up for a SaaS tool with their work email, connects it to the CRM, and suddenly customer data is flowing through a system that neither the IT department nor the data protection officer has had a hand in.
That isn't illegal in itself. But it is exactly the situation in which companies risk breaching the data protection rules without knowing it, because the attention has been on functionality, not on where data travels. The question "is it legal" is therefore rarely a yes/no question. It is a question of whether the right preconditions are in place.
What GDPR actually says about AI and customer data
GDPR applies to the processing of personal data, regardless of the technology. AI tools that read, analyse or generate text from information about identifiable customers or contact persons must be assessed under these rules. Customer data without personal data does not, for that reason alone, fall under GDPR.
That means the familiar GDPR principles apply in full when you put AI on customer data:
- Lawfulness, fairness and transparency: you need a basis for the processing, and customers must be able to understand what happens to their data.
- Purpose limitation: data collected for one purpose (e.g. support) may not simply be reused for another (e.g. training a model).
- Data minimisation: the AI tool should only have access to the data it actually needs.
- Storage limitation: data must not stay in the AI system longer than necessary.
- Integrity and confidentiality: you must be able to document that data is protected along the way.
The Danish Data Protection Agency has collected the rules and guidance on data protection on its website, and that is where you will find the up-to-date Danish interpretation of how the principles are to be applied in practice; see the Danish Data Protection Agency's rules and guidance.
The AI Act comes on top, not instead
It is worth noting that the EU's AI Act is a separate set of rules that imposes requirements on the AI system itself, particularly for high-risk uses. It does not replace GDPR. When you use AI for customer data, you typically have to deal with both sets of rules at the same time: GDPR for the data processing, the AI Act for the system's own risk classification and requirements. The European Commission's citizens' portal for businesses gives an overview of how the data protection rules fit together with digital legislation; see the EU's guidance on data protection for businesses.
Legal basis for processing: the first thing you need to have under control
Before the AI tool touches a single piece of customer information, you must be able to answer one question: what lawful basis do you have for processing this data in this way? The most common bases in a customer context are:
- Contract: the processing is necessary to perform a contract with the customer (e.g. AI answering support enquiries about an existing order).
- Legitimate interest: you have a legitimate business interest that is not overridden by the customer's interests and rights (e.g. internal AI analysis to improve response times).
- Consent: particularly relevant if data is to be used for something the customer could not reasonably expect, e.g. their enquiries being used to improve an AI model across customers.
The decisive point is that the basis must fit the specific use, not the technology in general. A legal basis that covers "answering customer enquiries" does not automatically cover "using customer enquiries to train an AI model". They are two different purposes, and the purpose limitation in GDPR means you must assess them separately.
The data processing agreement (DPA) with the AI vendor
When an AI vendor processes personal data on your behalf, there must be a data processing agreement. If the vendor also uses the information for its own purposes, that processing and role must be assessed separately. The European Commission's explanation of the roles describes the difference.
A DPA with an AI tool should as a minimum clarify:
- The purpose of the processing: precisely delimited, not just "delivery of the service".
- Whether data is used for model training, and if so, whether it can be opted out of, and whether that applies by default or requires active opt-in.
- Sub-processors: which third parties the AI vendor itself uses, and whether you have approved them.
- Deletion: what happens to data when you terminate the agreement or delete a conversation/case.
- Security measures: encryption, access control, logging.
Many AI vendors have standard terms written to cover as many customers as possible, not necessarily to protect your customer data in the best possible way. It is therefore worth reading the AI terms and the data processing agreement in concrete detail, rather than assuming that "they have it under control" because the product is professional.
Transfers of data to a third country
Clarify where the AI vendor and its sub-processors process the information, including remote access from third countries. Transferring personal data outside the EU/EEA requires a relevant transfer basis, for example an adequacy decision that covers the recipient and the transfer, or standard contractual clauses with the necessary supplementary measures.
In practice, that means you should be able to answer:
- Where are the servers that process your customer data physically located?
- Which transfer basis does the vendor use if data leaves the EU/EEA?
- Can you choose an EU-hosted version of the tool, if one exists?
It isn't illegal to use an American AI vendor, but it requires that the transfer has a legal foundation and that you have documented it. The Danish Data Protection Agency's guidance material on international transfers is the place to start if you are in doubt about a specific set-up.
Data minimisation and transparency towards customers
One of the most common mistakes is to give the AI tool access to the whole customer database because it is technically easiest, rather than limiting access to what the task actually requires. An AI tool that is to write draft replies to support enquiries doesn't need access to the entire sales history, payment details or internal notes about the customer.
Ask yourself: which fields, conversations and documents does this specific AI function really need? The narrower the access, the smaller both the legal risk and the consequence if something goes wrong.
Transparency towards customers
Customers have the right to know how their data is processed, and that also applies when part of the processing happens via AI. You should make sure that your privacy policy genuinely reflects that AI is involved in the handling of customer enquiries, and that it is clear where the customer can turn if they have questions or want to object.
Information about direct AI interaction is regulated separately in Article 50(1) of the AI Act, which places a duty to inform on the provider, with the stated exceptions. The data protection law requirements on information about the processing of personal data must be assessed alongside this.
Deletion and storage of data in AI systems
AI tools tend to accumulate data in logs, chat list histories and "memory" features that are easily forgotten in the ordinary deletion routine. When you introduce an AI tool, you should clarify from day one:
- How long does a conversation or case stay in the system before it is deleted or anonymised?
- Can you delete a single customer's data across the AI system if the customer asks for it (the right to erasure)?
- Are there copies of data in logs or cache that ordinary deletion doesn't reach?
This is often the point where AI tools differ most from classic systems: because data can lie scattered across several layers (chat history, embeddings, logs), and a single "delete" button doesn't necessarily reach them all.
Own instance and data ownership: what the architecture doesn't decide
The legal assessment requires an overview of the whole supply chain and of the processing that takes place at each link. The number of vendors, or the choice between shared infrastructure and an own instance, doesn't on its own determine the risk. Also document access, purpose and any transfers.
This is where architecture plays a part. An own instance can separate your application and database from other customers'. The number of processors and data flows depends on the specific set-up. The language model may, for example, run at a sub-contractor, and it has to be included in the agreements. Data processing agreements and any transfers must be assessed for the whole supply chain.
It is important to be precise here: an own instance is not a legal guarantee in itself, and it doesn't exempt you from making the same assessments of legal basis, purpose and deletion. Separating the application and database can limit sharing between customers. Risks in processing at sub-contractors must be assessed separately.
How to get started: a practical checklist
Whichever AI tool you are considering or already using, the order is typically the same:
- Map which AI tools already touch customer data in the company, including those individual employees have signed up for themselves.
- Identify the legal basis for each specific use, not just for the tool in general.
- Clarify the roles and enter into a data processing agreement with the vendors that process personal data on your behalf.
- Clarify the transfer basis if the vendor processes data outside the EU/EEA.
- Limit access to what each AI function actually needs.
- Set deletion rules and check that they also reach logs and chat history.
- Update the privacy policy so that it reflects the actual use of AI.
- Document the assessment to an extent that fits the processing and the risks. Clarify separately whether an impact assessment or other assessments are required.
Connection to your wider AI maturity
Using AI lawfully on customer data is closely linked to how mature your overall AI and data set-up is. If you already have your CRM data hygiene under control, for example, it is easier to delimit what an AI agent should have access to; see our review in the CRM data hygiene guide 2026. If you are considering more broadly what an AI agent in the company actually is and what it can be used for, it is worth starting with What is an AI agent for the business. And if you use AI in customer service or customer success specifically, there is a dedicated review in the Customer Success AI guide 2026.
What the AI should NOT do
- The AI agent should not itself make the legal judgement about the legal basis; that is a human assessment, possibly with the help of a lawyer or adviser.
- The AI should not have broader data access than the specific task requires, just because it is technically simpler to give full access.
- The AI should not be used to circumvent consent requirements or the right to object by "moving" the processing into a system the customer hasn't been informed about.
- The AI should not replace a required data processing agreement or the assessment of the parties' roles.
- The AI should not be used to make final decisions about customers without human oversight, where it has a legal or similarly significant effect on them.
Frequently asked questions
Q: May we use a free AI chatbot to answer customer enquiries?
It depends on the terms of the specific tool. Free versions of AI tools in some cases use data for model training as part of their business model, and they don't always have a data processing agreement sufficient for commercial use on customer data. Always check the terms and whether there is a paid/business version with a proper DPA before you put customer data on a free tool.
Q: Do we have to ask for consent before AI may process customer data?
Not necessarily. Consent is only one of several possible legal bases in GDPR. Often a contract or legitimate interest can be the right basis, depending on the specific processing. The decisive point is that you have actively chosen and can justify the basis, not that consent is always required.
Q: May an AI vendor use our customer data to train its model?
Model training must have a lawful basis and be compatible with the other data protection requirements. An agreement between you and the vendor, or an opt-out, is not in itself sufficient. Clarify the purpose, roles, information to customers and the vendor's actual use of the information.
Q: What if the AI vendor has servers in the US?
That isn't automatically illegal, but it requires a valid transfer basis, for example standard contractual clauses (SCCs). You should be able to see which transfer basis the vendor uses, and ideally have it documented in the agreement.
Q: Do an own instance and self-host automatically make it legal to use AI on customer data?
No. An own instance and self-host can give more control over operations, but the number of processors and any transfers depend on the specific set-up. You still need a legal basis, a deletion plan and the other GDPR preconditions in place; architecture doesn't replace the legal assessment.
Q: Can a customer demand that their data is deleted from an AI system?
The right to erasure also applies in AI systems when the conditions are met; it has exceptions. Clarify the scope of the request and deal with relevant copies in chat history, logs and other layers in accordance with the rules. The Danish Data Protection Agency's guidance on deletion also describes the handling of backups.
Situations where AI touches customer data, and what is typically required
| Situation | What is required | Why it is relevant |
|---|---|---|
| AI chatbot answers customer enquiries via chat/email | A specific legal basis; DPA if the vendor is a processor | Purpose, necessity and the parties' roles must be clarified |
| AI analyses customer data to provide insight or recommendations | A separate legal basis for the analysis purpose + data minimisation | Analysis is often a different purpose from the original collection |
| The AI vendor is hosted outside the EU/EEA | A valid transfer basis, e.g. SCCs | Transfer to a third country is regulated separately in GDPR |
| The AI tool can use data for model training | Clarify the lawful basis, purpose, roles and terms for training | An opt-out or a DPA doesn't in itself make the training lawful |
| A customer asks to have their data deleted | Assess conditions and exceptions; carry out relevant deletion | Affected copies and system layers must be part of the assessment |
Overall assessment: AI can be used legally for customer data, but it depends on the specific processing. Clarify the purpose, data types, basis, roles, security, information, storage and any transfers. An own instance doesn't change these obligations; the whole supply chain must be included. Use the checklist to find the questions your specific assessment must answer.
This article is editorial guidance and does not constitute legal advice. Contact your data protection adviser or the Danish Data Protection Agency if you have specific doubts about your set-up.
Tags: GDPR, AI and law, customer data, data protection, compliance
Contents
Unsure about the rules for your own set-up?
We look at what you use today and what it requires: documentation, consent and what the EU AI Act means for you. 20 minutes, no commitment.
Book a reviewRelated articles

MFA: multi-factor authentication for your business – how to get started
Learn how to implement MFA securely in your organisation. Guidance on methods, costs and best practices for protecting company data.

English-language invoices: requirements for invoices to foreign customers
Invoice in English to foreign customers legally. Danish regulations, EU and international VAT rules, zero-rating, reverse charge, and automation.

Public Procurement in Denmark: how to bid for government contracts
How Denmark's Public Procurement Act works for suppliers. Threshold values, tender procedures, exclusion grounds and AI solutions for proposal management.