Article 4 of the EU AI Act asks providers and deployers to ensure a sufficient level of AI literacy among the staff who use AI systems. Training that counts is role-appropriate, tied to the systems people actually use, and documented. A generic video watched once is weak evidence. This post is not legal advice.
#Key takeaways
- Article 4 applies to deployers as well as providers, so most companies using AI tools are in scope.
- The obligation has applied since 2 february 2025, according to the text of the act.
- Training should match each role and the AI systems that role actually uses.
- Keep records: who was trained, on which system, with what content, and when.
- For a legal reading of your own situation, ask your counsel; this post explains the practical side.
#What does article 4 of the EU AI Act require?
The EU AI Act, article 4 says providers and deployers of AI systems shall take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff and of other people dealing with AI systems on their behalf. It asks them to consider technical knowledge, experience, education and training, and the context the systems are used in.
The same page notes that article 4 has applied since 2 february 2025. The text does not prescribe a course, a number of hours or a certificate. That leaves the company to decide what sufficient means for each role, and to be able to show the reasoning.
#What kind of training actually counts?
Three features make training defensible, in our reading of the text:
- role-appropriate: a recruiter screening cvs with AI needs different knowledge from a developer building a model, or a finance analyst drafting commentary.
- tied to systems actually used: training covers the tools in your inventory, their limits, and where a person must check the output.
- documented: attendance, content, the system covered and the date are recorded, and refreshed when the tools change.
Generic awareness sessions can be a starting layer, but on their own they say little about how a given team uses a given system. The NIST AI risk management framework, january 2023 makes a similar point from the risk side: people overseeing AI need to understand the system they are accountable for.
#How should a company choose an AI training provider for article 4?
French buyers often ask which training organisation to choose for the article 4 obligation. Ask each provider four questions. Can they train by role rather than in one large group? Will they use your actual tools and documents? Do they produce records you could show an auditor? And are the trainers practitioners who use these systems at work?
A provider that answers yes to all four gives you training and evidence. One that offers a single certificate for everyone gives you a certificate.
We are
Stennir trains role cohorts on the AI systems your staff actually use and documents who was trained, on what, and with which content.
We aren't
Stennir is not a law firm, does not give legal advice, and does not sell a one-size certificate.
#How does Stennir approach AI literacy training?
The AI training service runs the AI Operator Program as role-specific cohorts on the real workflow. Each cohort starts from an inventory of the systems that role uses, practises on those systems, and ends with an adoption report that also serves as a training record. Usage is measured before and after, so the report shows change, not only attendance.
Training runs in English and Arabic, and French is available on request for teams in france.
#What should a compliance or L&D lead do next?
List the AI systems in use and the roles that use them; that inventory is where your training plan starts. Then book a 30-minute call or take the free readiness assessment. See also AI for compliance and our guide to choosing an AI training company.
#Common questions
Is this legal advice?
No. This post explains how Stennir reads article 4 and how we design training around it. It does not assess your company's legal position. For that, speak to your counsel or data protection officer, and bring them the inventory of systems and roles, which makes their review faster and more precise.
Does article 4 apply if we only use tools like Copilot or ChatGPT?
The article covers deployers, meaning organisations using AI systems under their authority, not only those building them. Companies using third-party assistants are generally in scope. The practical step is the same either way: train the people who use those tools, on those tools, and keep a record of it.
What records should we keep?
Keep, for each person, the role, the AI systems covered, the content of the session, the trainer, and the date. Add refresher records when tools change. Stennir's adoption report gathers this for each cohort, along with usage measured before and after, so the evidence and the outcome sit in one document.
Can the training be delivered in French?
Yes, on request. Stennir supports English and Arabic fully and runs other languages, including French, when a client needs them. The materials use your own tools and documents in the working language of each team, so staff practise on the systems and text they handle every day.
