AI literacy training is now a paperwork problem, not a training problem

The EU AI Act's Article 4 — AI literacy — is the obligation almost everyone overlooks, because it doesn't ban anything. It says providers and deployers must ensure people using or affected by AI systems have "a sufficient level of AI literacy." No fine schedule, no dramatic prohibition. Just a duty to train and, implicitly, to be able to show you did.

That makes it a paperwork problem. The question isn't "did we train people" — it's "who, and what do we keep as proof." So I built a kit around exactly those two questions.

Who needs training is the first output

Art. 4 doesn't require training everyone in the company. It targets the people who use the system, operate it, or are affected by its outputs — and the level scales with the risk of the system. The kit takes your setup and returns a scoped list: these roles touch the model, these operate it, these are affected. That scoping is the part teams get wrong by either training nobody or training everyone and proving nothing.

The risk-tier logic matters because literacy for someone prompting a chatbot is different from literacy for someone deploying a high-risk system. The kit keeps the level attached to the role instead of handing you one generic course.

What evidence you keep is the second output

Training without a record is training you can't prove. The kit outlines what to retain — who was trained, on what, when, and at what level — so that if asked, you have a trail rather than a memory. I'm explicit that this is the decision-support half: it tells you what to capture, it doesn't generate the certificates or store them for you.

The decision-support framing is deliberate. AiSkillKit helps you map the obligation to your roles and your evidence; it does not certify that you've satisfied Art. 4, because that depends on facts about your systems and people the tool can't see.

The limit I state up front

This is not legal advice and not an audit. The AI Act's literacy duty interacts with your specific systems, contracts, and member-state expectations in ways a generator can't resolve. I keep the copy honest: it organizes the thinking and the evidence list, you own the actual training and the proof.

What it also can't do is assess whether your AI system is high-risk in the first place — that's a separate classification step. The literacy kit assumes you already know what you're running; it tells you who to teach about it.

When a client asks who needs training, AiSkillKit maps the roles and the evidence in a pass, and I take it from there. It organizes the obligation; the actual training and proof stay with the human.

I've also started keeping the literacy scoping next to the system classification, because the two questions — what is this system, and who must understand it — are really one decision split across two documents. Treating them together is the only way the evidence trail stays honest instead of drifting apart over a quarter.

FAQ

Who does Art. 4 require training?
Not everyone. It targets people who use, operate, or are affected by an AI system, at a level scaled to the system's risk. The kit scopes that list from your setup.

What evidence should I keep?
Who was trained, on what topic, at what level, and when. The kit outlines the record; you maintain the actual proof.

Does a clean kit mean I satisfy Art. 4?
No. It's decision-support that maps the obligation to your roles and evidence. Satisfaction depends on your real systems and people, which needs human review.

Is this legal advice or an audit?
Neither. It's a planning aid for the literacy obligation. For a formal position, consult counsel familiar with the AI Act.