Article 4 AI Act

What Article 4 of the EU AI Act asks employers to do

Plain-English guide to AI literacy measures and training records under Article 4 — and what a short orientation can (and cannot) do.

Article 4 is the part of the EU AI Act that focuses on AI literacy at work. In plain terms, organisations that provide AI systems or put them into use are expected to take measures that support AI literacy among staff (and others) who deal with AI on their behalf — and to be able to show what those measures were. It is not a requirement to buy an official certificate. It is about suitable measures and evidence.

Key dates

AI literacy obligations under Article 4 have applied since 2 February 2025. On 27 July 2026 the “Digital Omnibus” (Regulation (EU) 2026/1744) rewrote Article 4 in the form described below. Other parts of the Act apply in phases — check the European Commission’s Q&A for the dates that matter to you.

What a “measure” is

A measure is a practical step that helps people who use AI at work do so more safely and responsibly. Since 27 July 2026, Article 4 says providers and deployers “shall take measures to support the development of AI literacy” of their staff, and that this “does not require” them “to guarantee any specific level of AI literacy of any individual” (Regulation (EU) 2026/1744). For most small professional-service teams, that starts with a shared baseline: how tools like ChatGPT or Copilot can go wrong, what not to paste into them, and when to escalate.

What a training record is

A training record is dated evidence that a named person completed a defined measure. The EU does not issue an official “AI literacy certificate” for staff. A clear individual record is evidence you took a measure — not proof of full AI Act compliance and not a government credential.

What a short orientation can do

A short orientation can give everyone who uses everyday AI tools a shared baseline: privacy and data handling, hallucinations, accountability, and escalate triggers. Think of it like a safety induction — serious enough to matter, short enough that people finish. Owners get a baseline done; they do not get “finished” with deeper governance.

What it cannot do

It is not legal advice. It is not training for high-risk AI systems. It does not make an organisation “Article 4 compliant” by itself, and it does not replace role-specific or legal review where those are needed.

Good enough for small teams?

Finish rate matters. Long LMS modules and free national courses can be excellent content — and still sit incomplete. For small agencies and consultancies, a measure people actually complete, with a record per person, is often more useful than a syllabus nobody finishes.

How Tovia fits

Tovia is about 10 minutes in WhatsApp, €20 per person + VAT, with an individual dated training record. It supports Article 4 with documented evidence for everyday, lower-risk professional AI use. It is not a compliance guarantee.

When to go deeper

If you operate high-risk systems, need role-deep expertise, or need legal interpretation of your obligations, use longer courses, workshops, or advisers. Tovia is the foundational layer — not the whole stack.

Questions teams ask

Is there an official EU AI literacy certificate?

No. The EU does not issue an official staff AI literacy certificate. What matters is whether you took suitable measures and can show them — for example with individual dated training records.

Does a 10-minute course make us “Article 4 compliant”?

No. Do not treat any short course as a compliance guarantee or as full AI Act compliance. Tovia is a practical measure with evidence support — not legal advice and not a claim that you are finished with Article 4 or the wider Act.

What’s the difference between a training record and a certificate?

A training record is evidence that a person completed a defined measure on a date. A “certificate” often sounds like an official credential. Tovia provides an individual dated training record (QR-verifiable). It is not an official or government certificate.

Is this for high-risk AI systems?

No. Tovia is a foundational layer for everyday tools like ChatGPT and Copilot. High-risk systems need deeper, role-specific training and governance.

We’re a small agency — do we still need a measure?

If your people use AI at work on your behalf, Article 4 literacy expectations can still apply. Size doesn’t remove the idea of suitable measures — it changes what is proportionate. A short completed baseline with records is often a sensible start for small teams.

How is this different from free national or EDIH courses?

Many free courses are strong on content. Tovia sells completion in WhatsApp plus a clear individual record for the team — so the measure is finished and filed, not only assigned.

Can training happen on WhatsApp?

Yes, if the flow is clear and you keep a proper record per person. Tovia runs in WhatsApp and issues a dated training record you can verify later.