All posts

EU AI Act Article 4: what AI literacy asks of an owner-led business

Article 4 asks any business using AI at work to support its team's AI literacy. What the amended text says, who it covers and what to keep on file.

Natan Trolka · Founder, Otomatic Digital

The short answer

Article 4 of the EU AI Act requires providers and deployers of AI systems to take measures to support the AI literacy of their staff and of others using AI on their behalf. A deployer is any business using an AI system under its authority. The duty has applied since 2 February 2025. No certificate is required, and no set level is mandated for any one person.

Your team probably uses AI at work already: a chat assistant for drafts, an AI feature inside the CRM, a note-taker on calls. Article 4 of the EU AI Act puts a duty on exactly that, and it covers businesses that use AI as well as the ones that build it.

It sets no course, certificate or required level, and since 2 August 2026 national authorities can enforce it.

What Article 4 says

Providers and deployers of AI systems must take measures to support the development of AI literacy among their staff, and among other people who operate or use AI systems on their behalf.

The measures should take account of three things:

  • the technical knowledge, experience, education and training of the people involved
  • the context the AI systems are used in
  • the people or groups the systems are used on

The wording changed on 27 July 2026, when Regulation (EU) 2026/1744 amended the Act. The 2024 text asked businesses to ensure "to their best extent, a sufficient level of AI literacy". The amended text asks for measures that support it, and states that no specific level has to be guaranteed for any individual.

Does it apply to your business

The Act defines a deployer as a person or business using an AI system under its authority. If your staff use AI tools for their work, your business is a deployer of those tools, whoever built them.

The European Commission's own Q&A gives the example of employees who use ChatGPT to write advertising copy or translate text: they should be told about specific risks, such as the tool inventing facts.

An owner asked me whether they legally need this. Yes: a business whose staff use AI has to take measures to support their AI literacy, and training is the usual one. I compare it to fire safety training: nobody waits for the fire to learn the exit. Untrained staff can paste private information into a chatbot before anyone notices.

Article 4 sits in Chapter I of the Act, which has applied since 2 February 2025. Supervision and enforcement by national market surveillance authorities started on 2 August 2026. The Commission describes that enforcement as proportionate, taking account of factors such as whether an infringement was intentional or negligent. In Ireland, no single body is named for Article 4: S.I. No. 405 of 2026 splits market surveillance across sector regulators, such as the Data Protection Commission, the Central Bank and the Workplace Relations Commission, and designates the AI Office of Ireland (Oifig IS na hÉireann) as a market surveillance authority alongside them.

What counts as a measure

The Commission is clear that no certificate is needed. It suggests organisations can keep an internal record of training and other guidance, and that measures should at least give staff a general understanding of AI, its opportunities and its risks, matched to their role and the tools they use.

The mistake I see most often is staff not knowing what happens to what they type. On personal accounts of ChatGPT, Claude and Gemini, chats can be used to train the provider's models, depending on a setting in each account, and Google says human reviewers read a subset of Gemini chats.

On the business plans (ChatGPT Business and Enterprise, Claude for Work, Gemini in Google Workspace), each provider says it does not train on your content by default. So the rule for staff is simple: no client names, addresses or bank details in an AI tool the business has not approved.

For an owner-led business, we suggest four things:

  • a list of the AI tools in use, who uses them and for what
  • a short usage note: what information stays out of AI tools, and who checks an output before it reaches a client
  • training on the tools each role actually uses, with practice on real tasks
  • a record of who attended which session and when

That list is our recommendation. Your obligations depend on what you use AI for, so take legal advice where a use is sensitive.

Where training fits

Training is the measure most businesses reach for first, because it covers the general understanding and the role-specific practice in one go.

It also deals with the fear. Most staff who worry about AI do not know how to use it or what it can do, and the pace of change unsettles them.

Some worry about security, with reason: IBM's 2025 Cost of a Data Breach Report found that one in five organisations reported a breach due to shadow AI, meaning AI tools staff use without approval.

Our AI Training & Workshops run as a 90-minute introduction, a 3-hour workshop or a training day, on site or online. Your team practises on its own tasks and keeps the session materials, which also gives you the record. We do not issue a certificate, because the law does not ask for one.

This post is general information, not legal advice.

Sources

  1. 01EUR-Lex, Regulation (EU) 2024/1689, consolidated text including Regulation (EU) 2026/1744, Articles 3(4), 4 and 113, version of 27 July 2026.
  2. 02EUR-Lex, Regulation (EU) 2024/1689, original text, Article 4, 12 July 2024.
  3. 03Official Journal of the EU, Regulation (EU) 2026/1744 (Digital Omnibus on AI), Article 4 on entry into force, published 24 July 2026.
  4. 04European Commission, AI Literacy: Questions & Answers, updated 20 October 2026.
  5. 05Irish Statute Book, S.I. No. 405 of 2026, European Union (Artificial Intelligence) (Designation) (Amendment) Regulations 2026, Regulation 2(c) and Schedules 1 to 3, made 31 July 2026.
  6. 06OpenAI Help Center, How your data is used to improve model performance, read 16 November 2026.
  7. 07Anthropic Privacy Center, consumer plans and commercial products, read 16 November 2026.
  8. 08Google, Gemini Apps Privacy Hub and Generative AI in Google Workspace Privacy Hub, read 16 November 2026.
  9. 09IBM, 2025 Cost of a Data Breach Report announcement, 30 July 2025.

Frequently asked

Ready to remove the work itself? Start with Workflow Automation.

Start here

Find the work your week repeats.

Tell us your business priorities and bottlenecks in a 30-minute call. Our initial recommendations, the fit decision and the next step follow in writing, and they are yours whether you hire us or not.

Starts with a 30-minute Business Check and your AI Recommendation Plan.