AI competence according to Article 4 AI Act: Training as an operational process
By Nico Freitag, Geschäftsführer, Cybersecurity und Governance
Article 4 of the AI Act requires providers and operators to take measures to ensure a sufficient level of AI competence among people working with AI systems. A general one-off training course is not automatically sufficient. Content and evidence must take into account the role, prior knowledge, operational context, affected people and specific risks and be updated when changes occur.
Who does the competency requirement affect?
The European Commission explains that providers and operators of AI systems must take measures for employees and other people who are involved in operation or use on their behalf. This doesn’t just apply to development and IT. Business users, purchasing, management, support or external service providers can also make decisions that change the quality, data use or effects of a system. An inventory therefore connects every AI application with roles and actual tasks.
How do meaningful learning goals differ by role?
Users must master permissible entries, limits, control obligations and reporting channels. Specialists need methods for quality testing and approval. Technology requires knowledge of data flows, access, model statuses, logging and secure operation. Leadership and purchasing must classify roles, risk, vendor commitments and responsibility. A common basic block can convey concepts and attitude, but is supplemented by role-related exercises.
What content should be included in a practical AI training course?
Works with the actual released systems and anonymized cases from the company. Practices good job description, reviewing evidence, recognizing uncertainty, handling personal or confidential information, and confident rejection. Also discusses when a human decision remains mandatory and how errors are reported. The focus is not on as many prompt tricks as possible, but on safe and technically usable work.
How is sufficient competence demonstrated?
Documents target group, learning objectives, content, appointment, participation and practical success control. A short knowledge question can test concepts, but not the safe handling of the process. Therefore, use case tasks, observation or a moderated exercise. Record open points and measures. The evidence should show why the format was appropriate for the role and context, not just that a file was opened or participation was confirmed.
When does the learning concept need to be updated?
New systems, significantly changed models, additional data types, new roles, incidents or changed legal guidelines trigger an audit. Complements short learning impulses in everyday life and repeats critical exercises. The Commission describes its sample collection as a living resource. This fits in with a competency process that incorporates operational and incident experiences rather than sending out the same slides every year. Legal boundary issues should be examined professionally.
From the answer to implementation
Related service
View the scope, delivery model and responsible contacts.
Open →Related product
See a practical product path connected to this topic.
Open →Working checklist
Prepare the next decision with a structured checklist.
Open →All specialist articles
Continue with reviewed answers from the same practice areas.
Open →Sources and basis
The central statements in this article were reviewed against the following primary sources.
Frequently asked questions
- Do all employees have to receive the same AI training?
- No. The scope and depth should be based on prior knowledge, role, operational context and risks. A common basic block can still make sense.
- Is a certificate of participation sufficient proof?
- It demonstrates participation, but does not automatically demonstrate appropriate competence. Learning objectives, content, role reference and practical success monitoring strengthen the evidence.
- Does the obligation only apply to high-risk AI?
- Article 4 refers generally to providers and operators of AI systems. The specific measure takes into account the context and those affected.