AI Law, Risk & Governance

AI Literacy under Article 4 of the AI Act: What the Training Obligation Requires and How to Implement It

Marin Niehues21 September 202612 min read

Marin Niehues gesturing with a headset and name badge during a talk on stage.

Article 4 of the AI Act requires every company that provides AI systems or uses them professionally to take measures that build the AI literacy of its staff. The obligation applies regardless of risk class, so it also covers everyday use of a language model at work. Since its revision by the Digital Omnibus regulation, the provision no longer demands a guaranteed level of competence per person, but it does demand measures that fit the use and can be evidenced. This article explains what the obligation covers, how a role-based program is structured and how documentation stands up to scrutiny. This article provides an overview and does not constitute legal advice.

What does AI literacy mean under Article 4 of the AI Act?

Under the AI Act, AI literacy means the knowledge and understanding that allow people to deploy AI systems in an informed way and to be aware of the opportunities and risks of AI. The definition sits in Article 3(56) of Regulation (EU) 2024/1689, the obligation in Article 4. The definition expressly includes the possible harm AI can cause. What it describes is judgment in working with the system, beyond mere operation.

The provision sets out what the measures should reflect. On the one hand, staff's prior knowledge and experience. On the other, the context in which the systems are used and the people they affect. A clerk drafting text with a language model needs different knowledge from a team reviewing AI recommendations for credit decisions. The yardstick is the role in the process, and the program follows that distinction.

Is AI training mandatory under the AI Act?

Yes, measures to build AI literacy are mandatory for providers and deployers of AI systems, although the regulation prescribes no particular format. Training is the obvious form, supplemented by guidance or supported practice. In its questions and answers on AI literacy, the European Commission notes that merely pointing to instructions for use is generally not enough.

Two groups are covered. A deployer is any company that uses an AI system under its own authority in a professional capacity, with only purely personal use excluded. Almost every company that uses language models such as ChatGPT or AI features in business software therefore falls under the obligation. A provider is anyone who develops an AI system or places it on the market under its own name. The measures address the company's own staff and other persons working with the systems on its behalf, such as service providers and freelancers.

The regulation as a whole, with its risk classes and roles, is covered in EU AI Act: what companies need to know.

What the Digital Omnibus revision changed

The Digital Omnibus regulation (EU) 2026/1744 replaced Article 4 and clarified the obligation on companies. The original text required providers and deployers to ensure, to their best extent, a sufficient level of AI literacy among their staff. The new text obliges them to take measures that support the development of AI literacy. It states expressly that no specific level of AI literacy of any individual has to be guaranteed.

Two further paragraphs were added. The Commission and the member states support providers and deployers, in particular SMEs, in meeting the obligation, and the Commission publishes practical examples on its single information platform. The European AI Board may adopt recommendations based on European competence frameworks.

In practice, the obligation remains and its object is the measure. Companies must show that they planned and delivered suitable measures. A test result per person is not required. Documentation therefore gains weight, because it is the evidence that the measures fit the use.

Supervision and consequences of missing AI literacy

Compliance with Article 4 is supervised by the national market surveillance authorities, which in Germany under the national implementing act is generally the Federal Network Agency, and BaFin for the financial sector. The AI Act itself sets no separate fine for breaches of Article 4. The Commission points out, however, that missing training and guidance weigh particularly heavily once an incident has occurred.

The real risk therefore lies in the event of damage. If an employee enters confidential data into a publicly available language model or adopts a flawed output unchecked, supervisors and business partners ask what precautions were in place. A documented literacy program is then the evidence of proper organization. High-risk systems add a separate duty: deployers assign human oversight to persons with the necessary competence and authority.

The figures show that the obligation touches a real bottleneck. A lack of knowledge slows adoption more than any other factor. A well-built literacy program therefore addresses two tasks with one measure, the legal duty and the ability to use AI well.

The tier model: building AI literacy by role

An effective program distinguishes four tiers by role in the process and assigns content and scope to each. The assignment follows from the AI inventory: who uses which system, and what effect do its outputs have on people and decisions?

The tiers build on each other. Everyone completes tier 1, and further tiers follow from the role. For the executive team, tier 3 is part of the core program, because it decides on deployment and approval. How a program for managers is structured is shown in AI training programs for executives. Tier 1 content is closely tied to the internal AI policy for employees, since it sets out the rules the training conveys.

How do you document AI literacy for the AI Act?

AI literacy is documented with an internal record of target groups and the measures completed per person, and no certificate is required. The Commission states in its questions and answers that internal records suffice. An external certificate is permissible, but it does not replace evidence that the content fits the systems in use.

The record should be easy to evaluate. A learning management system or a maintained spreadsheet suffices, as long as status and gaps are visible at a glance. The decisive point is the link to the inventory: when a new system is added, its approval automatically triggers training for the affected roles.

Yes, the legal department is affected twice, as a user of its own AI tools and as the function that supports compliance with the AI Act across the company. Lawyers who review contracts with AI or research with language models belong at least in tier 2. On top of that comes the knowledge needed to assess use cases from other units legally.

The legal department's program therefore includes confidentiality and professional secrecy, along with liability questions and the verification of sources in AI outputs. In many companies, legal also signs off the Article 4 documentation in substance. Steering the program, by contrast, belongs to the line that is accountable for AI use, so that training and rollout run together.

Implementation in four steps

Companies implement Article 4 in four steps, which in most cases can be completed within a quarter. The first step is the inventory of systems in use, including AI features in purchased software. The second assigns roles to tiers. The third creates and delivers the content, starting with tier 1 for all users. The fourth anchors evidence and updates in the approval process for new systems.

In Germany, the works council has co-determination rights over the implementation of in-company training. Involving it early in selecting content and participants avoids delays. Federal Employment Agency funding is not available for the legally required basic training, although it is for advanced qualifications. How training is embedded in the rollout of an AI application is described in the article From pilot to scale.

Niehues Consulting as a partner for AI literacy in operation

Article 4 can be handled as a training task or solved as part of the rollout, and the second route delivers the greater benefit. Niehues Consulting takes the second route from a delivery perspective. AI management produces the inventory and portfolio of applications from which target groups and tiers are derived. During rollout, roles and processes are adjusted so that training and evidence work in daily operations alongside the approval of new systems.

The approach follows the Niehues Delivery Model with the phases Understand, Build, Roll out. Three of its review questions matter most for AI literacy. Who decides, who delivers, who uses? Are users enabled and processes adapted? Which metric shows after 90 days that the solution is landing? Marin Niehues brings experience from projects ranging from VC-backed scale-ups to DAX 40 corporations in 20+ countries on six continents. A free 30-minute initial consultation, which you can arrange via the contact form, is the natural first step.

Frequently asked questions

Does AI Act training need to be certified?

No, the AI Act requires no certificate for AI literacy training. The European Commission states that internal records of measures and attendance suffice. A certificate does not replace evidence that the content fits the systems in use.

Does the obligation apply if staff only use ChatGPT?

Yes, the obligation applies to any professional use of AI systems, regardless of risk class. A company that provides ChatGPT or Microsoft Copilot to staff, or permits their use, is a deployer. The Commission gives the example of informing staff about risks such as hallucinations.

Does Article 4 apply to contractors and freelancers?

Yes, besides the company's own staff, the obligation covers persons working with its AI systems on its behalf. That includes freelancers and service providers. In practice, a contractual clause confirming relevant knowledge plus access to the tier 1 materials covers this group.

How often must AI literacy training be repeated?

The regulation sets no fixed cycle, but the measures must fit current use. A refresher is therefore due with every new system that brings a new user group, plus an annual review of content. Linking training to the approval of new systems keeps the program current without a separate calendar.

What is the penalty for breaching Article 4 of the AI Act?

The AI Act sets no separate fine for Article 4, and supervision lies with the national market surveillance authorities. The greater risk arises in the event of damage, when missing measures are treated as an organizational failure. A documented program is then the most important evidence.

Who is responsible for AI literacy in a company?

Responsibility lies with the executive team, which delegates it to a named person. In practice that is the function steering the AI portfolio, supported by learning and development and the legal department. What matters is a single accountable owner for program and evidence.

What must AI literacy training cover at a minimum?

Basic training explains how the AI in use works, including its typical errors, and conveys the rules of the internal AI policy. The Commission also recommends clarifying the company's role as provider or deployer and the risks of the specific systems. For users in business units, reviewing outputs in their own process is added.

Let’s talk about your initiative.

In 30 minutes we clarify where your organization stands and which step will pay off most now.

Request a consultation

30 minutes · free · no obligation