AI Act training (Article 4): the AI literacy obligation for companies. What actually applies in 2026

Article 4 has applied since 2 February 2025. Providers and deployers of AI systems must, to their best extent, take measures to ensure a sufficient level of AI literacy among staff and other people operating or using those systems on their behalf. Measures should reflect role, knowledge, experience, risk, and context. We run training from €1,200 net per day.

A lot of noise has gathered around this article. One training day and three documents do not automatically close the compliance question. A program, attendance list, and materials can show what an organization did. The organization must still assess whether those measures fit the system, people, risk, and context, and whether competence is maintained as work changes.

Quick answer

• What applies: Article 4 of the AI Act, the duty to take measures toward sufficient AI literacy among staff and other people operating or using AI systems on a provider's or deployer's behalf.
• Since when: since 2 February 2025. This is an obligation that is already active, not a future deadline.
• Who it covers: providers and deployers of AI systems. In practice, that includes many companies using AI professionally.
• What it means in practice: measures suited to risk and context, such as training, instructions, exercises, operating rules, and knowledge checks. Records evidence those measures; they do not automatically prove compliance.
• What it costs with us: AI training (Claude) for companies from €1,200 net per day. The team offer is on the team training page.

Who does Article 4 cover, and since when?

Article 4 speaks about providers and deployers of AI systems. In practice this means almost every company that uses AI at work: a law firm that summarizes filings, an accounting office that classifies invoices, an e-commerce shop with a sales assistant, an HR team that processes CVs, a marketing team that generates content.

The duty rests with the provider or deployer and concerns the competence of its staff and other people who operate or use AI on its behalf. People or groups affected by the system influence the assessment of context and risk, but that alone does not make them a group subject to the training duty. Competence should match the role: one level for a team using a ready-made tool, another for someone designing an AI-based process.

The date is unambiguous: 2 February 2025. Anyone waiting for the rule to "come into force" is waiting for something that has already happened.

What is at stake if you skip training?

Here you have to be precise, because this is where fear-selling usually appears. The honest answer: Article 4 itself does not carry its own fine tariff. There is no provision that says "no training equals X in penalties".

That does not make the obligation meaningless. Staff competence is examined through market surveillance, and its absence feeds the assessment of a company's due diligence when an AI-related incident occurs: a wrong decision, a data leak, a discriminatory outcome. A company that cannot show it took care of people's competence stands on weaker ground: in an inspection, in a dispute, in an assessment of liability.

So we treat Article 4 as a real documentation obligation, not as a box to tick and not as an axe over your head. The point is not to dodge a mythical fine. The point is to hold proof that the company acted reasonably.

Which AI Act deadlines actually apply in 2026?

This is the only table you need. The rest is noise.

The postponements come from a package known as the Digital Omnibus: political agreement was reached on 7 May 2026, the European Parliament approved it on 16 June 2026, and publication in the Official Journal is still under way (as of July 2026). The two dates that matter most to companies, Article 4 and Article 50, remain unchanged.

An important caveat: we give the high-risk obligations as market context, not as a date that every company must put in its calendar. Most ordinary uses of AI in a company are not high-risk. Do not take on a deadline that does not apply to you.

For the full picture, four facts about when the AI Act applies (as of July 2026):

• Article 4 (AI literacy): applies since 2 February 2025.
• Obligations for general-purpose AI (GPAI) models and governance rules: apply since 2 August 2025.
• General, full applicability of the regulation: 2 August 2026.
• Obligations for high-risk systems: transition periods run to December 2027.

For an average company using AI at work, the first two lines are what bind, above all Article 4. The rest is context, not a calendar to copy over.

What does training that supports the obligation look like?

Training under Article 4 is not a lecture about the future of AI. Its job is twofold: to genuinely raise the team's competence and to leave behind a trail that attaches to the company's documentation.

Our AI training (Claude) for companies is run by engineers with systems in production, not staff trainers. The difference is practical: the day goes to participants' real tasks, to what may and may not be handed to AI, where the boundaries begin, and how to recognize a result you cannot trust. That is AI literacy in the sense the rule intends, not a set of curiosities.

The day divides well when the group is matched by level rather than gathered "everyone at once". A team that is just starting needs the basics: what AI does well, where it errs, how to phrase an instruction, and how to check the result. People who already use AI need something else: repeatable workflows, data-safety rules, and the limits of autonomy. That split is, in fact, exactly what Article 4 calls for, a level of competence adequate to the role, not one training for everyone.

After the day, the client is left with three artifacts that can enter its Article 4 records:

• a training program matched to the team's roles and level,
• an attendance list of participants,
• the materials and exercises from the workshop.

Those artifacts do not determine compliance on their own. The organization must assess whether the program fits participant roles, system risk, and work context, and when competence needs to be refreshed.

To date we have delivered five one-on-one training sessions. We run the team format on the same program. The price starts at €1,200 net per day; the full terms are on the team training page, and all service lines with prices are on the pricing page.

Disclaimer: our materials are input to the client's documentation, not legal advice. Classification of AI systems and operational duties stay with the company. We provide the competence and the trail; the decision on how to fit them into a compliance policy is the company's, with its own lawyer where needed.

What does Article 4 NOT require?

Most of the fear comes from things that are not in the rule. A few myths to delete:

• It does not require a certificate from an accredited body. Article 4 says nothing about an official exam or a stamped document. What counts is real competence and the company's ability to show it.
• It does not require identical training for everyone. The level should match the role. Someone who summarizes a document once a week needs something different from someone designing a process with AI in it.
• It does not require registration or notification to an authority. For ordinary use of AI there is no duty to report anywhere. The documentation stays with you in case of an inspection.
• It does not require mandatory annual repetition on a fixed date. Competence has to be maintained when the technology or the process changes, but the rule does not impose a rigid training calendar.
• It does not introduce its own monetary penalty. As above: Article 4 is a duty to keep people competent, not a schedule of fines.

If you hear an offer built on these myths, you are hearing fear being sold. The real obligation is narrower, more concrete, and entirely manageable.

FAQ

Since when does Article 4 of the AI Act apply?

Article 4 has applied since 2 February 2025. It requires providers and deployers of AI systems to take measures toward a sufficient level of AI literacy among the people covered by the rule.

Are there fines for not running AI Act training?

Article 4 itself does not set a specific fine tariff. That does not mean it can be ignored. Competence supports safe AI use, and an authority may assess the measures an organization took, including after an incident.

Who does the AI literacy obligation cover?

The duty rests with providers and deployers of AI systems. It covers their staff and other people who operate or use AI systems on their behalf. People or groups affected by an AI system inform the risk and context assessment, but that alone does not make them a group that must be trained.

Are a program, attendance list, and materials enough for Article 4 compliance?

Not automatically. They can evidence the measures taken, but adequacy depends on the organization's role, system risk, people's knowledge and experience, and the context of use. The organization should review competence and update its measures when the system or work changes.

Who must take AI Act training?

Article 4 does not prescribe one mandatory course for everyone. A provider or deployer should tailor measures to staff and other people who operate or use AI on its behalf. Scope depends on role, knowledge, experience, risk, and context.

Is AI training mandatory in 2026?

Article 4 requires measures that support sufficient AI literacy, but those measures do not always have to be a single course. They may include workshops, instructions, operating rules, exercises, and periodic knowledge checks tailored to role, risk, and context.

Where to start

If you first want to understand where AI even sits in your company and how to treat it, a map of the concepts helps: what is an AI agent explains the difference between a chatbot, a copilot, and an agent, because the level of autonomy shapes what competence the team needs.

If you already know you need training with a real Article 4 trail, the next step is simple:

1. Request team training and tell us who uses AI in the company and for what.
2. We match the program to roles so the day goes to participants' real tasks, not generalities.
3. After the training you get the program, attendance list, and materials as evidence of the measures taken. Your organization assesses their adequacy and adds any other controls required by its risk and context.

Request team training | See pricing | What is an AI agent

Related articles

• What is an AI agent: the seven criteria
• AI agent: how much implementation costs in Poland in 2026
• Syntalith