Testamur.
Regulation (EU) 2024/1689 · Article 4

Your people watched the training. Can you show they learned anything?

Article 4 of the EU AI Act requires organizations to support a sufficient level of AI literacy among staff. It has applied since February 2025. National authorities began enforcing it in August 2026. And it reaches companies with no European entity at all.

1 Aug 2024
AI Act enters into force with a phased schedule.
2 Feb 2025
Article 4 begins to apply. The duty is live from this date, not from 2026.
2 Aug 2026
National market surveillance authorities begin supervising and enforcing Article 4.
Ongoing
Literacy gaps weigh as an aggravating factor in enforcement of other obligations.
The obligationArt. 4

What Article 4 actually says

Providers and deployers of AI systems must take measures to support a sufficient level of AI literacy among their staff and other people operating AI systems on their behalf, taking into account their technical knowledge, experience, education, training, and the context the systems are used in.

That is close to the whole of it. It is one of the shortest obligations in the regulation and one of the broadest, because it applies regardless of whether the AI you use is high risk, limited risk, or minimal risk. If your finance team uses a copilot to draft analysis, that is in scope. If your recruiters use screening software someone else built, that is in scope.

Two things follow from the wording that most organizations miss. The duty reaches contractors and anyone operating AI on your behalf, not only employees. And it is explicitly proportionate to role and context, which means a single organization-wide briefing is unlikely to satisfy it for staff whose work is materially affected by AI output.

Territorial scopeArt. 2

Why a US company with no EU office can be in scope

The AI Act does not follow where you are established. It follows where your AI is used and where its output lands. Providers and deployers based outside the EU are bound where the output produced by their AI systems is used within the union.

In practice that catches a lot of American mid-market companies who assume this is somebody else's problem:

Common routes into scope

If this is trueThen
You have EU customers whose users receive AI-generated outputIn scope as a provider or deployer depending on your role
You employ anyone based in an EU member state, including remote staffThose people are staff for the purposes of the obligation
A partner embeds your AI output into a service sold in EuropeReach can be indirect and still count
You screen candidates, including EU applicants, with automated toolsIn scope, and additional obligations may apply on top

The corollary matters too. If none of these are true today, Article 4 does not apply to you today. One European client changes that, usually without anyone noticing.

The standardUndefined

Nobody will tell you what sufficient means

This is the part that makes Article 4 genuinely awkward rather than merely administrative. The regulation mandates no curriculum, no certificate, and no examination. There is no approved course, no accreditation body, and no score to hit.

That flexibility is a gift and a trap. You get to design an approach that fits your organization. You also have to decide what sufficient means, and be able to show your reasoning if anyone asks. Nobody is coming to validate your answer in advance.

It is also worth being precise about what is not required, because vendors in this space are not always careful about the distinction. There is no legal obligation to measure your employees' AI literacy. Recording what training has been undertaken carries clearer expectations. Measurement is not the requirement. It is the strongest evidence available that the level you chose was in fact sufficient, and that it improved.

MisreadingsFive common

What organizations get wrong

“We have until August 2026.”
No. The obligation has applied since February 2025. What began in August 2026 is supervision and enforcement, not the duty itself.
“There must be a required certification.”
There is not. No curriculum, certificate, or exam is mandated. Any vendor selling you a compulsory accreditation is selling you something the regulation does not ask for.
“We sent an all-hands email about AI.”
Thin. A single generic communication is unlikely to amount to a proportionate, role-based effort, and it produces almost no evidence.
“It only applies to our employees.”
Wider. The obligation explicitly reaches other people operating AI on your behalf, including contractors and temporary staff.
“Enforcement of Article 4 alone is unlikely, so we can wait.”
Half right, and the wrong half. Standalone enforcement is expected to be uncommon. The exposure is that a documented literacy gap sits alongside any other investigation as an aggravating factor, which raises your risk across everything else.
EvidenceWhat holds up

What you would actually have to produce

Since the standard is yours to define, the question is not whether you did enough. It is whether you can show what you did, why you judged it sufficient, and what it achieved. Most organizations have the first, sometimes. Very few have the third.

Weak positionDefensible position
Training was made available to anyone who wanted itTraining assigned by role, based on how each team actually uses AI
Completion records scattered across an LMS, Slack, and inboxesOne record: who, what, when, mapped to role
Completion rates reported as the outcomeA measured baseline showing what people understood before and after
Nobody named as accountableA named owner, with the reasoning behind the chosen standard written down
Done once, at rolloutRe-measured as AI use in the organization changes

The gap between the two columns is not usually effort. It is that most organizations bought training and assumed the evidence came with it.

Eight questions. An honest answer about where you stand.

Find out whether Article 4 applies to your organization and what your current evidence position would look like if someone asked. Takes about three minutes.