AI literacy under Art. 4 of the AI Act
Providers and deployers have to ensure that the people dealing with their AI systems have a sufficient level of AI literacy. It is one of the few obligations where the question of role does not matter.
The duty is briefly worded and therefore often overlooked. It requires no certificate, but it does require a deliberate decision about who needs which knowledge, and evidence that the knowledge was conveyed.
What a workable implementation contains
- A review of who in the organisation deals with which systems
- A grouping by intensity of use rather than one training for everyone alike
- Content on the capabilities and limits of the systems in use
- Content on errors, bias and how to handle outputs
- Content on confidentiality, data protection and the internal rules
- Documented attendance and a cycle for refreshers
- A usage policy that the training refers to
Two points that are often confused
The duty binds the organisation, not the individual. Not every person has to know everything. What is required is that the necessary competence exists in the organisation as a whole and sits where it is needed.
Separate from that is human oversight of a high-risk system. Article 26(2) AI Act requires people with the necessary competence, training and authority for that role. That standard stands on its own and is not tied to Article 4, which is why general training does not satisfy it.
How we support you
We determine the group of people covered, develop the content along the systems you actually use and record the implementation so that it can be evidenced. On request we deliver the training ourselves.
AI in use, training outstanding?
We bring the duty to a level that can be evidenced.
Get in touchFrequently asked questions
Who exactly does the duty cover?
Your own staff and other people dealing with the systems on your behalf. Depending on the arrangement that includes service providers. It does not mean end customers who merely receive an output.
What level is required?
A sufficient level, measured against the knowledge, experience and training of the people involved as well as the context of use and the groups affected. For a team operating a high-risk system that means something different than for staff occasionally using an assistant.
Since when has the duty applied?
Since 2 February 2025. Article 4 AI Act sits in the first chapter of the regulation and therefore covers every use of AI systems, regardless of risk class. It is triggered as soon as staff use a generally available tool in ordinary office work.
What happens if we do not meet it?
Article 99 AI Act provides no separate fine for a breach of Article 4, so there is no direct sanction. Indirectly it still becomes expensive. Running a risk management system with staff unable to assess the requirements means Article 9 AI Act is not met, and that carries up to 15 million euros or three per cent of worldwide annual turnover. The civil side comes on top, because a lack of competence can be treated as an organisational failing.
How do we evidence it?
The regulation prescribes no format. What works in practice is a combination of training with documented content and attendance, a written usage policy and a refresher when things change materially. Without documentation, compliance cannot be shown afterwards.