Business

Training your staff in AI: what the AI Act really requires after the 2026 change

On this page
  1. What's going around, and why it's no longer accurate
  2. Does it affect my company?
  3. What Article 4 said, and what it says now
  4. What hasn't changed
  5. What it doesn't require
  6. What to do in your company
  7. And if I don't?
  8. You're not on your own: support for SMEs
  9. What I take away

What's going around, and why it's no longer accurate#

If you've looked up the AI Act (the EU's Artificial Intelligence Regulation) and staff training, you've probably read that, since February 2025, every company that uses AI must ensure its staff have a sufficient level of AI knowledge.

That roughly summed up the original wording. But in July 2026 the European Union amended the AI Act with the so-called Digital Omnibus on AI, Regulation (EU) 2026/1744, and the article that covers this, Article 4, changed. It's still easy to find guides that quote the old wording.

In this post I cover what Article 4 said, what it says now, what it still requires of you, and what you can do in your company. Everything comes from the text of the regulation and the European Commission's official answers, linked at the end.

Before you read on: what this post is and isn't

It's general information, not legal advice. I'm not a lawyer: I teach AI. Quotes from the law are word for word and come from Spain's official gazette (BOE) and the consolidated AI Act text as at 27 July 2026 published by the European Commission; everything else comes from the Commission's official answers. All the sources are linked at the end.

It's up to date as of 13 October 2026. These rules are changing: the Spanish law is still going through Parliament and the Commission keeps publishing guidance. Before making decisions, check whether anything has changed and take your specific case to a legal professional.

Does it affect my company?#

Probably. The AI Act distinguishes between whoever develops an AI system (the provider) and whoever uses it (the deployer). Article 3 defines a deployer as any person or organisation, public or private, that uses an AI system under its own authority, except when it's used for a personal, non-professional activity.

In other words: if people in your company use AI for work, your company is a deployer. You don't have to develop anything. The Commission makes this clear with an example in its questions and answers: a company whose employees use ChatGPT to write advertising copy or translate texts is covered by Article 4, and those employees should be informed about its risks, such as hallucinations.

What Article 4 said, and what it says now#

1 Aug 2024 The AI Act enters into force 2 Feb 2025 Article 4 starts to apply 26 May 2026 Spanish draft law 27 Jul 2026 Text with the new Article 4 Aug 2026 Supervision begins
Fig. 1 — Article 4 has applied since February 2025, changed in July 2026, and national authorities have supervised it since August 2026.

The original 2024 wording said that providers and deployers:

"shall take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff […]"

And this is the current wording, after the Omnibus (paragraph 1):

"Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf, taking into account their technical knowledge, experience, education and training and the context the AI systems are to be used in, and considering the persons or groups of persons on whom the AI systems are to be used. This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual."

The difference, in short:

Before (2024)Now (2026)
What you have to doTake measures to ensure a sufficient levelTake measures to support AI literacy
Level required"Sufficient", undefinedNone in particular, for anyone
Public supportNot in the articleThe Commission and Member States must help, especially SMEs

Paragraphs 2 and 3 are new. The Commission and the Member States must support companies, in particular SMEs, and the Commission will publish practical examples of how to comply. The AI Board, for its part, will adopt recommendations with common objectives.

And what exactly is "AI literacy"? Article 3 defines it as the skills, knowledge and understanding that allow an informed use of AI systems and an awareness of their opportunities, their risks and the harm they can cause. It isn't knowing how to build AI: it's knowing how to use it with good judgement.

In one sentence

The obligation still exists, but it no longer asks you to guarantee a specific level: it asks you to take real measures, tailored to who uses AI and what for.

What hasn't changed#

The change softens the obligation, but doesn't remove it. Several things stay the same:

  • It's still an obligation on companies. In November 2025 the Commission had proposed moving it to the Member States and the Commission itself, but the adopted text keeps it on providers and deployers.
  • It has applied since 2 February 2025. Article 4 is in Chapter I of the regulation, and Article 113 sets that date for Chapters I and II. According to the Commission, the obligation already applies.
  • Someone is already supervising it. Not the European AI Office, but the national market surveillance authorities, since August 2026.
  • If you use high-risk AI, the bar will be higher. Article 26 requires human oversight of those systems to be assigned to people with "the necessary competence, training and authority", and the Commission points out that this obligation remains in place. It will apply later, though: after the Omnibus, the high-risk rules apply from 2 December 2027 for Annex III systems (such as those used in employment or education) and from 2 August 2028 for those embedded in Annex I products.

What it doesn't require#

The Commission's questions and answers settle several common doubts:

  • No certificate is needed. An internal record of training sessions and other initiatives is enough.
  • You don't have to measure each employee's knowledge, or guarantee a specific level.
  • You don't need an "AI officer" or any specific governance structure.
  • There's no mandatory course format. The Commission insists there's no one-size-fits-all. It does warn, though, that just relying on a tool's instructions for use, or asking staff to read them, might be ineffective in many cases.

What to do in your company#

The Commission suggests some minimum steps. This is my practical version for an SME:

  1. Take stock. Which AI tools are used, who uses them and what for? Include the ones people use on their own, like free assistants in the browser.
  2. Be clear about your role. Do you only use AI developed by others (deployer), or do you also develop it (provider)? Almost every SME is in the first group.
  3. Identify the risks. Made-up answers that look true, personal or confidential data ending up in a public tool, decisions about people taken without human review…
  4. Tailor the training to each profile. Someone who drafts emails with an assistant doesn't need the same as someone who uses AI to screen CVs. Article 4 itself asks you to consider knowledge, experience and the context of use.
  5. Write a usage policy. Which tools are allowed, which data never goes in, and when output must be reviewed before it's used.
  6. Keep a record. Dates, content and attendees of each training or session. That's the internal record the Commission talks about.
Fun fact

The Commission applies Article 4 to its own staff: an internal framework of basic AI competences, training packages by profile (generalists, managers and specialists), an internal portal, regular Q&A sessions and a network of "AI Champions". It says so in its own questions and answers.

And if I don't?#

According to the Commission, national authorities can impose penalties and other measures for breaches of Article 4, under each Member State's laws. Those penalties must be proportionate and take into account, case by case, the nature and gravity of the breach and whether it was intentional or negligent. It also warns that a penalty might be more likely if there's proof of an incident due to a lack of appropriate training and guidance.

In Spain, the Council of Ministers approved the draft Organic Law on the good use and governance of artificial intelligence on 26 May 2026 and sent it to Parliament. The draft designates the supervisory authorities, including AESIA (the Spanish Artificial Intelligence Supervisory Agency), and sets out the penalty regime. As of this post it is still a draft: it doesn't appear in the BOE, Spain's official gazette, as an adopted law.

You're not on your own: support for SMEs#

The new paragraph 2 requires the Commission and the Member States to support companies, especially SMEs. Today there are already:

  • the Commission's questions and answers on AI literacy, which are kept up to date;
  • a repository of practices with more than 40 real initiatives from companies and public bodies (the Commission notes that copying them doesn't on its own guarantee compliance);
  • the network of European Digital Innovation Hubs (EDIHs), more than 200 across Europe, which help SMEs and public bodies digitalise. According to the Commission, 9 out of 10 already offer AI services, including training.

What I take away#

  • Article 4 has been softened, not removed. The obligation to take measures still stands, and it's already supervised.
  • What it asks is reasonable: know which AI your people use, what for, with what risks, and train them accordingly.
  • Complying takes work done and documented, not certificates: an inventory, tailored training, a usage policy and a record.
  • Watch out for old guides. If they cite a mandatory "sufficient level", they're quoting the 2024 text.
Personal note

Since 2024 I've been teaching AI applied to business, to teams and to professionals, and one of my courses starts precisely from this obligation. With the 2026 text in hand, my conclusion is that the answer to "do I have to train my staff?" is still yes, but the approach changes: it isn't about passing an exam, it's about people knowing how to use the tools they already have well.

References

  1. Reglamento (UE) 2024/1689 de Inteligencia Artificial (texto original)BOE · boe.es
  2. Regulation (EU) 2026/1744 (Digital Omnibus on AI)EUR-Lex · eur-lex.europa.eu
  3. AI Act, consolidated text as at 27 July 2026EUR-Lex · eur-lex.europa.eu
  4. Artículo 4: Alfabetización en materia de IAAI Act Service Desk (Comisión Europea) · ai-act-service-desk.ec.europa.eu
  5. AI Literacy - Questions & AnswersEuropean Commission · digital-strategy.ec.europa.eu
  6. Living repository to foster learning and exchange on AI literacyEuropean Commission · digital-strategy.ec.europa.eu
  7. AI Act Service Desk: FAQEuropean Commission · ai-act-service-desk.ec.europa.eu
  8. El Gobierno aprueba el proyecto de ley que garantizará una supervisión humana y un uso confiable de la IAMinisterio para la Transformación Digital y de la Función Pública · digital.gob.es
  9. Real Decreto 729/2023, Estatuto de la AESIABOE · boe.es

Share

Keep reading

Business

8 signs your company website has a security problem

Eight checks any business owner can do in a few minutes, with no technical knowledge, to find out whether their website has a security problem, and what to do if they find one. With the official source for every figure.

· 8 min

Business

What AI can automate in your business (and what it can't)

A practical map of what a business can automate with AI, department by department: what AI adds to a traditional automation, what's best left to a person, and how to spot where to start.

· 8 min

Want to train your team in AI?

I design hands-on AI training for teams and managers, tailored to your sector and the tools you already use, with materials and follow-up afterwards.

Let's talk→