AI School
Training by role with records and evidence, and in-house trainers to sustain it.
See AI SchoolResources · Ironhack for Business
Does your company meet the mandatory AI training duty? Take the test and find out.
Five questions to see whether you meet the AI literacy duty in the European regulation, what evidence you are missing, and in what order to close it. With no invented fine figures.
Why this matters
What protects you is not the course. It's the record.
Article 4 of the European AI regulation requires that the people operating AI systems have a sufficient level of literacy, matched to their role, the context of use and the risk of what they do. It has applied since February 2025 and is supervisable from 3 August 2026.
You will meet plenty of people selling you a fine figure. We are not going to, because Article 4 does not appear in the regulation's penalty regime: the €35m and €15m amounts belong to other articles, and in Spain the law that will set the penalties is still going through parliament. What is real is more uncomfortable: if there is an AI incident at your company tomorrow and you cannot show you trained whoever was using it, you have no defence. The exposure is not the fine. It is the absence of evidence.
Outcome
Knowing where you stand
A clear coverage status and the list of what is still undocumented.
Certainty
From the primary source
Based on the text of the regulation, not on a vendor's talking points.
Time
Two minutes
Five questions answered yes, no or partly.
Effort
No audit needed first
Answer with what you already know about your company today.
Test
Five questions about Article 4
Is anyone at your company using AI tools in their work?
Is there a written AI usage policy that has been communicated to staff?
Have you trained the people who use AI, according to their role and risk level?
Do you keep evidence of who received that training and when?
Have you defined which tools are approved and which are banned for company data?
Coverage status
Unanswered
Requirements covered—
Supervisable from3 Aug 2026
What Article 4 requires, and who supervises it
It requires providers and deployers of AI systems to ensure a sufficient level of literacy among the staff operating them, taking into account their knowledge, the context of use and the people affected.
The duty has applied since February 2025. Supervision sits with national authorities and starts on 3 August 2026. It must be proportionate, and action is more likely where there is an incident attributable to a lack of adequate training.
The first question does not score: it decides whether the article applies to you. The other four are the four pieces you have to be able to show.
What it requires
What your company has to be able to demonstrate
Article 4 is written in terms of outcome, not procedure: it requires a sufficient level of literacy, but does not say how many hours of training or with what syllabus. That unsettles a lot of people, though it is really an advantage: it lets you fit it to your company. What it does not let you do is improvise it the day somebody asks. In practice, complying comes down to four pieces.
A written usage policy
A document saying what can and cannot be done with AI at your company, with which data, and who decides in borderline cases. Without it, each team applies its own judgement and there is nothing to audit or to show. It is the cheapest of the four pieces and the one most people skip.
Training matched to role and risk
The regulation asks for a level sufficient for what each person does. Someone using an assistant to draft emails does not need what someone connecting a model to customer data needs, or what someone deciding which tool gets bought needs. A single course for the whole workforce complies badly and wastes budget doing it.
Documentary evidence of who was trained and when
This is the piece people forget and the only one that genuinely defends you. A course delivered with no record proves nothing six months later. You need to know who attended, what, on what date and with what content, and you need to be able to show it without reconstructing it from memory.
And it has a practical consequence almost nobody anticipates: the evidence expires the moment new people join. If you trained the workforce in March and by September you have brought in fifteen people, your coverage is no longer what it was. With no onboarding process, this decays on its own.
A list of approved tools
Knowing which tools are authorised, which are banned, and what kind of data can be used in each. It is the point where the literacy duty meets data protection, and where the surprises usually turn up: people uploading confidential information to services nobody has reviewed.
Where to start if you have nothing
The order with least friction is policy first, tool list second, training by role next, and record-keeping from day one. Trying to start with the training before defining what is allowed produces courses that contradict the company's actual practice.
Frequently asked questions
Frequently asked questions about Article 4 of the AI Act
From exactly when does it apply?
The Article 4 duty has applied since 2 February 2025. There is no grace period: if your company uses AI, it already applies to you.
What changes on 3 August 2026 is supervision: from that date national market surveillance authorities can act.
Is there a fine for not training my staff?
There is no specific amount attached to Article 4 in the European regulation. The penalty regime lists which articles carry which amounts, and Article 4 is not on that list: the €35m or 7% belongs to prohibited practices, and the €15m or 3% to obligations of providers and deployers of high-risk systems.
Member states set their own regimes for the rest. In Spain, the AI governance law that will develop the penalties is still going through parliament after being approved in the Council of Ministers.
If someone shows you a calculator that gives you a fine figure for not training your team, that figure is invented.
Which companies does it apply to?
Providers and deployers of AI systems. In practice, if your company uses AI tools in its operations, you are a deployer and it applies to you.
It does not depend on company size or sector. It depends on whether anyone uses AI in their work.
What training counts as sufficient?
The regulation sets no hours and no syllabus: it asks for a level appropriate to the person's prior knowledge, the context the system is used in, and the people it affects.
That means the right answer differs by role, and that a single generic course for the whole workforce complies worse than differentiated tracks, even shorter ones.
How do I demonstrate compliance?
With documentation. A communicated usage policy, tracks by role, a record of attendance and content, and a list of approved tools.
The practical test is simple: if you had to show your compliance to a third party tomorrow, could you do it with what you already have on file, without fabricating anything? If the answer is no, that is your gap.
Does this training qualify for funding?
Yes. AI training qualifies through the programmed training system, so a good part of the cost can come out of the credit your company already has set aside.
It is the uncommon case of a regulatory duty that arrives with part of the budget already paid.
Does this replace legal advice?
No. This test is a self-assessment tool based on the text of the regulation and is for working out where to start.
For your specific case, and especially if you operate in a regulated sector or use AI in decisions affecting people, consult your legal advisers.
Next step
Complying is a project, not a one-off course.
Training the workforce is the visible part. What is needed underneath is a usage policy, a list of approved tools and a record that survives a review. That gets built in the diagnostic, and comes out with an owner and a date.
Related resources
Your FUNDAE credit
How much of that mandatory training can be paid for with the credit you already have.
Calculate the creditHow many hours your team gets back
The other side of the argument: beyond complying, how much time AI returns.
Open the calculatorThis is not legal advice. Article 4 of Regulation (EU) 2024/1689 has applied since February 2025, and its supervision, in the hands of national market surveillance authorities, starts on 3 August 2026. This test deliberately does not return a penalty amount: Article 4 is not among the provisions with a specific amount in the regulation's penalty regime, and in Spain the rule that will develop the penalties is still going through parliament. For your specific case, consult your legal advisers.