AI The European AI Act became fully applicable to high-risk systems on 2 August 2026. If your company uses artificial intelligence for anything that affects people, this concerns you, even if you did not develop it yourself.
The date that matters
On 2 August 2026 the obligations for high-risk AI systems came fully into force, together with the transparency obligations of Article 50 and the oversight of general-purpose models. Since that date, the Spanish authority can inspect and impose penalties.
The fines are not symbolic: they reach 35 million euros or 7% of global turnover, whichever is higher. And the scope is extraterritorial: if you operate in the European market, it does not matter where you are headquartered.
What the AI Act is, in two sentences
It is the European regulation governing artificial intelligence. It also classifies each use according to its risk.
The four risk levels
Prohibited
Uses the regulation does not allow under any circumstances. If your project falls here, there is no version “with warnings” that can save it.
High risk
Systems that make decisions about people: employment, credit, education, access to essential services. These carry the most obligations.
Limited risk
Basically, transparency. Conversational assistants and most uses in a small business fall here.
Minimal risk
Everything else. No specific obligations, although it is worth documenting what you use and what for.
Where a typical small business fits
This is the reassuring part. If you use AI to draft texts, classify emails, summarise documents or answer questions with an assistant, you are most likely at limited or minimal risk.
However, “limited” does not mean “nothing”. The main obligation is transparency: anyone talking to a machine has the right to know it is a machine, and generated content must be identifiable as such.
The AI Act in a small business, step by step
First, make a list of the AI tools you use. Then note down what each one is used for.
What you do have to do, even if you are small
- Inventory which AI tools are used in the company, including the ones someone signed up for on their own.
- Classify each use by risk level: most will end up at the bottom, but you have to check.
- Tell people when they are dealing with an automated system, in the chat itself and not in the small print.
- Document what data goes in and comes out, because this overlaps directly with the GDPR.
- Appoint someone as responsible, even part-time: without an owner, this does not get maintained.
Shadow AI
The most common risk is not a wrongly declared high-risk system. It is the tool someone on the team started using without telling anyone, with customer data inside it.
That is why the inventory is the first step and not the last. You cannot comply with a rule about something you do not know exists, and most companies discover when doing it that they use more AI than they thought.
Training and the AI Act
The regulation requires that people who use AI know what they are doing. That is why basic training for the team is a good idea.
How we approach it
When we set up artificial intelligence for businesses, the notice that there is a machine behind it and the record of what is processed are part of the job, not a later phase. It is cheaper to do it this way than to add it once it is already in production.
And when AI connects to other systems, systems integration is exactly the point where it is decided which data leaves from where. It is best that someone makes that decision, rather than letting it happen by itself.
The AI Act and the GDPR are not the same
They are often confused, and it is worth separating them because they require different things. The GDPR deals with personal data; the AI Act deals with systems, whether or not they involve personal data.
So you can comply with one and not the other. A system that classifies warehouse parts does not touch personal data, but it is still an AI system; and an assistant that answers customers is subject to both at once.
- The GDPR asks: what personal data you process, on what legal basis and for how long.
- The AI Act asks: what system you use, for what decision and at what risk level.
- When both apply, the documentation overlaps quite a lot, so it is worth doing it once and doing it well.
AI providers and the AI Act
Ask your providers how they comply. That way you will know which part is your responsibility and which is theirs.
A timetable to keep it manageable
None of this has to be solved in an afternoon. A reasonable order for a small company would be this:
First, the inventory of tools, which is done by asking the team. Then, the risk classification, which for most is settled in a one-page table. Next, the transparency notices where needed, which are usually two or three sentences. And finally, periodic review, because tools change on their own.
With that, a small business that uses the AI Act as a checklist rather than as a threat has the work done in weeks, not months.
In short
The AI Act does not prohibit using artificial intelligence: it requires you to know what you use, what for, and to say so when it affects a person. For most small businesses it is a matter of getting organised, not engineering.
If you would like to review what you are using and where it fits, we start there in the free strategy call.
Where to read the text of the AI Act
The official text is published on EUR-Lex. However, for day-to-day purposes it is enough to understand the obligations that apply to your case.
What not to lose sight of
The AI Act was not written to hold back small companies, but to set rules where decisions affect people. So the further your use is from that border, the less paperwork you have.
However, that can only be shown if you have it in order. A one-page inventory and a risk classification are, in practice, the difference between being able to answer an awkward question and having to improvise.



