Knowledge base
Do I need to worry about the EU AI Act?
For most small business automation, the answer is barely, as long as you build it properly. Integrations, dashboards and assistants with human oversight fall under light transparency requirements, not the strict high-risk regime. This is a plain explanation, not legal advice.
The short answer.
The EU AI Act is the European law that sorts AI by risk and attaches requirements to each level. It sounds heavy, but the weight sits in a few categories, such as AI in hiring, credit scoring or medical diagnosis. The everyday automation a small or midsize business deals with almost always falls into the lightest categories. You tell people they're talking to AI, and you keep a human responsible for what matters. At its core, that's all there is to it.
If you build with documentation and with human approval on everything that goes out, you meet the spirit of the law without a separate project. That's exactly how I build, so for my clients this is rarely an obstacle. The requirements will get stricter over the coming years, though. If you document your processes properly now, you won't have to play catch-up later.
The four risk levels, and where your AI usually lands.
The EU AI Act has four levels. Most small business automation sits at the bottom, at limited or minimal risk.
| Risk level | Example | What it means for you |
|---|---|---|
| Unacceptable | Social scoring, manipulation, deception | Prohibited. Doesn't come up in normal small business automation. |
| High risk | AI in hiring, credit, exams, medical diagnosis | Strict requirements. Rare in standard small business automation, but watch out with hiring and staff selection. |
| Limited risk | A chatbot or AI assistant that talks to customers | Transparency. Tell people they're talking to AI. Otherwise light. |
| Minimal risk | Reading invoices, dashboards, integrations, spam filters | Almost no extra requirements. This is where most of the work falls. |
What building it properly already covers.
Most of the requirements at the lighter levels are exactly what a good builder does anyway. Be open about AI being involved, so a customer knows they're talking to an AI assistant and not to a person. Keep a human responsible for decisions that matter, as with the review handling at Inge de Bruijn Zwembad, where a staff member approves before anything goes out. Stick to the data you really need, host within the EU, and make sure the model refers the question on when in doubt instead of guessing, so you keep the risk of a hallucination small.
Also, write it down. What does the system do, what data does it run on, and who checks it? That documentation pays off twice. You meet the spirit of the law, and if something happens to me, someone else can take over. That's why I deliver code and documentation in your name by default.
Honest about the limits.
Two things to keep in mind. First, if you do use AI in hiring, credit scoring or anything else that directly affects people, it can be high risk, and then heavier requirements really do apply. Second, this isn't legal advice. I build with transparency and human oversight baked in, but if you want to know for sure how the law applies to your specific case, take it to a lawyer. We'll simply cover it in the free process review, so you know whether it applies to you or not.
Frequently asked questions.
Does my AI automation fall under the EU AI Act?
Almost all small business automation falls under limited or minimal risk. That includes integrations, dashboards, document processing and assistants with human oversight. These mainly come with transparency requirements, not the high-risk regime. Only AI that directly affects people, such as in hiring or credit scoring, can be high risk.
What's the minimum I need to do to comply?
For the lighter levels, it comes down to being open about AI being involved, keeping a human responsible for important decisions, using only the data you need, and documenting what the system does and what data it uses. That's exactly how I build by default.
Does an AI chatbot have to say it's AI?
Yes. That's the core of the transparency requirement at the limited-risk level. A customer has to know they're talking to an AI assistant and not to a person. I set that up by default in the assistants I build, and it helps trust too, because being honest works better than pretending.
Is this legal advice?
No. I build with transparency and human oversight baked in, so you meet the spirit of the law, but for a definitive judgment on your specific situation, bring in a lawyer. In the free process review, I will tell you honestly whether the EU AI Act is something to watch for in your plans.
Not sure whether it applies to you?
In the free process review, I look at whether your plans fall under the light or the heavy requirements, and I build with transparency and human oversight baked in either way. So you know where you stand, without a separate legal project.