What applies in practice, and what needs to be written down

The biggest gap is seldom legal. It is that nobody has written down what applies.
Employees who do not know what they are allowed to do guess. And they guess generously — customer material gets pasted into a personal account because it was quicker. That is a concrete risk, long before any authority gets in touch.
Two sets of rules are relevant. GDPR applies as soon as you process personal data, and has done so since 2018 — AI does not change the rules, only how easy it has become to break them by mistake. The EU AI Act is the EU's regulation for AI systems, which entered into force in 2024 and is being introduced in stages, with requirements that depend on what the system is used for.
This page describes what that means in practice and what I help with. It is written to be understandable, not exhaustive.
About this content: I am an AI consultant, not a lawyer, and this is not legal advice. The information on the regulations is general, was written in August 2026 and was updated in November 2026 (the section on GDPR-safe AI). For questions that concern real liability — high risk levels, exercise of public authority, decisions that affect individuals — you should speak to a lawyer, and always check against the current guidance from IMY (the Swedish data protection authority).
The position as of November 2026
The question "is there GDPR-safe AI?" assumes an AI service that is approved in itself. There is no such seal. GDPR regulates how personal data is processed, not which tool is used, so the same service can be used lawfully in one context and unlawfully in another. "GDPR-safe AI" is therefore a way of using AI, not a product.
This is what you need to be able to answer for every use that involves personal data:
Legal basis and purpose. Which basis under Article 6 do you have for processing the data in this particular step, and is only what is needed for the purpose processed (Article 5)?
Roles and agreements. When a supplier processes personal data on your behalf, the supplier is normally a processor, and a data processing agreement is then required (Article 28).
Where the data is processed. If the data leaves the EU/EEA, for example to the USA, you need a basis for the transfer (Chapter V), such as the supplier being certified under the EU–US Data Privacy Framework or the use of standard contractual clauses. In November 2026 the Commission's decision on the framework applies; the General Court dismissed an action against it in September 2025, and the appeal (case C-703/25 P) is being heard by the Court of Justice and had not been decided when this was checked on 14 November 2026. A ruling there could change the position.
Training. Check whether what you enter is used to train the model. According to the European Data Protection Board's Opinion 28/2024 (December 2024), the use of an AI model can be affected if it was developed with unlawfully processed personal data, unless the model is anonymous.
Risk assessment and information. Processing that is likely to result in a high risk to individuals may require a data protection impact assessment (Article 35). Data subjects must be informed about the processing (Articles 13–14), and automated decisions with legal or similarly significant effects on individuals are restricted (Article 22).
Organisations with professional secrecy have a sharper situation. For law firms there is a separate walk-through of GDPR-compliant AI for legal practices.
GDPR
The cheapest measure is the last one in the left-hand column: choosing areas of use where personal data does not need to be involved. A draft of a product text needs no customer data. A compilation of your own routines needs no personal information. A great deal of the value lies in material that was never sensitive to begin with.
The EU AI Act
The most important thing to understand is the difference between building and releasing an AI system and using one that someone else has built. A business that only uses ready-made tools belongs to the second category and therefore has considerably fewer obligations.
The requirements then depend on what the system is used for. Ordinary office use — drafts, summaries, translations — is lightly regulated. Use that affects individuals in matters such as employment, credit assessment or access to services is regulated considerably more strictly, and there you need to speak to a lawyer rather than to me.
One requirement that applies broadly is that anyone who uses AI at work must take measures to support the AI competence of their staff. Since the omnibus of July 2026, no particular level is guaranteed, but the requirement remains. In practice it is a training requirement, and it is one of the reasons I would rather train everyone in the basics than a few people in the advanced material.
Another is transparency: if a chatbot is used with customers, it should be clear that it is a machine, and material created with AI may need to be labelled depending on the context.
Policy
A short list. Not "AI tools may be used with judgement", but the names of the services you have paid for and checked. Anything else requires someone to ask first.
Personal data about customers and employees, health information, confidential agreements, login details. A concrete list beats principles — principles are interpreted generously when things are urgent.
Everything that goes out to a customer, everything that becomes a basis for a decision, everything that is published. The rule should be simple enough to remember without looking it up.
A named person. Without one, the answer to every borderline case becomes either a no that nobody follows, or a yes that nobody has thought through.
A policy longer than that is seldom read, and a policy that is not read protects no one. I help to prepare it, go through your actual areas of use and hold a walk-through with the staff so that the policy becomes known. You receive a fixed quote after a first free meeting.
What the EU AI Act concretely requires of a small business, and which dates apply after the high-risk rules were postponed, is set out in the article the EU AI Act for small businesses. On the policy document itself: AI policy for small companies.
One route into the high-risk category for a small business is recruitment: AI in recruitment — where the rules actually bite.
For law firms, professional secrecy comes on top of GDPR, and a language model that invents case law is a risk of its own: AI for law firms.
Frequently asked questions
Not as a product that is approved in itself. GDPR concerns how you process personal data, so it depends on how the tool is used: that you have a legal basis, a data processing agreement, know where the data is processed and do not enter more than is needed. The position as of November 2026.
Not in an ordinary personal account. If personal data is to be processed in an AI tool, you need a business service with a data processing agreement, control over whether the material is used for training, and clarity about where the data is processed.
The simplest advice is not to put personal data into the steps where it is not needed at all.
A business that uses AI tools that someone else has developed has fewer obligations than the party that builds the systems.
The requirements depend on the use: ordinary office use is lightly regulated, whereas use that affects individuals in matters such as employment or credit assessment is regulated considerably more strictly.
Four things go a long way: which tools are approved, what must never be entered, when a person must review before something goes out, and whom to ask when something is unclear. A policy that is longer than that is seldom read, and a policy that is not read protects no one.
The scope decides, and you receive a fixed quote after a first free meeting. This covers a review of your actual areas of use, preparation of an AI policy that can be followed, and a walk-through with the staff so that it becomes known. I am not a lawyer and do not replace legal advice.
Yes. Anyone who uses AI at work must take measures to support the AI competence of their staff (Article 4 of the EU AI Act), and regardless of the regulation, it is the measure that in practice reduces risk the most.
No. I am an AI consultant and help with the practical side: mapping how you actually use the tools, writing a policy that can be followed, and training the staff.
For questions that concern real liability I refer you to a lawyer, and I would rather say so once too often.
A free half-hour call. We go through how you use AI today and what needs to be written down.
Book a first meeting