AI literacy under Article 4: what the EU AI Act requires

Article 4 of the EU AI Act on AI literacy has applied since February 2025. As amended by Regulation (EU) 2026/1744, since July 2026 it requires all companies that provide or deploy AI systems to take measures that support their staff’s AI literacy (as of October 2026 · not legal advice). Yet reality paints a sobering picture: according to a recent Fraunhofer study, only 22 percent of employees in Germany feel competent using AI tools. The gap between the legal requirement and day-to-day operations can become expensive for many companies.

This guide shows what Article 4 requires in concrete terms, where German companies stand today, and how to build an AI training program that meets the compliance requirements and brings real value to your organization.

What does Article 4 of the EU AI Act require?

Article 4 of the EU AI Act is titled “AI literacy.” The provision has applied since February 2, 2025, without a transition period; Regulation (EU) 2026/1744 amended it with effect from July 27, 2026 (as of October 2026 · not legal advice).

Core requirements of the provision

In the version now in force, providers and deployers of AI systems take measures to support the development of AI literacy of their staff and of other persons dealing with the operation and use of AI systems on their behalf. They take into account technical knowledge, experience, education and training, and the context of use. They do not have to guarantee a specific level of AI literacy for anyone. The previous version still required them to ensure a sufficient level of AI literacy.

In concrete terms, this means the following for companies:

• Who is affected: every company that provides or deploys AI systems, regardless of size. Whether you are a sole proprietor using ChatGPT or a corporation with its own AI department, the obligation applies to everyone.
• Scope: the measures must fit the role. A clerk who uses an AI-assisted word processing tool needs different knowledge than a data scientist who trains machine learning models.
• Records: the law prescribes no form. If you record which training measures have taken place, you can show the authority what you have done. An informal verbal briefing is hard to prove.
• Continuity: AI literacy is not a one-time seminar. Because AI technologies evolve quickly, training should be updated regularly.

Enforcement and liability

Neither the EU AI Act (Article 99) nor Germany’s KI-MIG (Section 15) provides a separate fine for violations of Article 4 (as of October 2026 · not legal advice). The provision still has consequences:

• Without AI literacy measures, the risk of violating obligations that carry fines rises, for example the deployer obligations for high-risk AI: up to €15 million or 3 percent of global annual revenue (Article 99(4)).
• Under the KI-MIG, in force since July 29, 2026, the competent market surveillance authority is the Federal Network Agency (Bundesnetzagentur), and BaFin (Germany’s Federal Financial Supervisory Authority) for the financial sector.
• Then there is liability risk: if an employee makes mistakes that harm third parties because of a lack of AI literacy, the company can be held liable for inadequate training.

Relationship to other obligations

Article 4 does not exist in isolation. It complements other obligations under the EU AI Act such as the transparency requirement (Article 50), documentation obligations for high-risk AI (Article 11), and the duty of human oversight (Article 14). AI literacy is essentially the foundational prerequisite for fulfilling all other obligations: Only those who understand AI can deploy it transparently, document it, and oversee it.

The status quo—where does AI literacy stand in Germany?

The legal requirement is settled. But what does reality look like in German companies? Current studies and surveys paint a nuanced but, on the whole, worrying picture.

Facts and figures

In early March 2026, the Fraunhofer Academy published a comprehensive study on AI literacy in Germany. The findings are alarming:

• Only 22 percent of employees feel competent in using AI tools.
• In the 50+ age group, this figure drops to just 11 percent—a particularly critical finding given the demographic structure of many midsize companies.
• At the same time, the Bitkom survey from March 2026 shows that 62 percent of German companies identify AI as the most important continuing education topic of the year.
• However, only 35 percent of companies already offer structured AI training.

These figures reveal a massive gap between awareness and implementation. Nearly two-thirds of companies know they need to act—but only one-third is already doing so.

The biggest barriers

Why is there such a wide gap between aspiration and reality? The Fraunhofer study identifies several key obstacles:

Fear of job loss (47 percent): Nearly half of employees cite the concern of being replaced by AI as the biggest barrier to their own continuing education. This result underscores that AI training must not only convey technical knowledge but also address change management aspects. People who are afraid do not learn effectively.

Lack of structures: Many companies, especially midsize ones, have neither a dedicated continuing education budget for AI nor clear responsibilities. Training often happens “on the side”—if it happens at all.

Unclear requirements: which measures are enough? The legislator deliberately leaves this open; the European Commission publishes practical examples (Article 4(2)). Each company still has to decide for itself what its employees need to know.

Rapid technological change: what counts as current knowledge in January 2026 may be outdated by July. The pace of the AI industry makes continuing education an ongoing task, and it overwhelms traditional training concepts.

Generational gap: With only 11 percent of those over 50 feeling AI-competent, midsize companies face a particular demographic challenge. Experienced professionals who possess enormous domain knowledge need to be inspired to use AI tools without feeling patronized.

The 3-tier competency model

In early March 2026, the IHK network (Germany’s chambers of industry and commerce) published a position paper recommending a tiered competency model. It defines different requirements for each role instead of treating all employees the same.

• Criterion: Target group | Tier 1: Foundation: All employees in the company | Tier 2: Practitioner: Employees who regularly use AI tools | Tier 3: Expert: AI developers, AI project managers, AI officers
• Criterion: Content | Tier 1: Foundation: AI fundamentals, terminology, opportunities and risks, EU AI Act basics, data protection fundamentals | Tier 2: Practitioner: Prompting, tool proficiency, quality assurance, workflow automation, everyday compliance | Tier 3: Expert: Model selection, fine-tuning, RAG architecture, AI governance, risk management, auditing
• Criterion: Duration | Tier 1: Foundation: 1 to 2 days | Tier 2: Practitioner: 3 to 5 days | Tier 3: Expert: 10 to 20 days
• Criterion: Cost per person | Tier 1: Foundation: €500 to €2,000 | Tier 2: Practitioner: €2,000 to €5,000 | Tier 3: Expert: €3,000 to €8,000
• Criterion: Outcome | Tier 1: Foundation: Basic understanding, increased awareness | Tier 2: Practitioner: Productive AI use in daily operations | Tier 3: Expert: Building and maintaining internal AI systems
• Criterion: Link to the EU AI Act | Tier 1: Foundation: Basis for measures under Art. 4 | Tier 2: Practitioner: In-depth training for staff of AI deployers | Tier 3: Expert: In-depth training for AI providers and high-risk systems

Which tier does your company need?

The answer depends on your AI usage:

• Standard AI tools only (ChatGPT, Copilot, AI-powered email sorting): Tier 1 for everyone, Tier 2 for power users.
• Custom AI workflows (n8n automations, RAG systems, AI-powered customer advisory): Tier 1 for everyone, Tier 2 for involved employees, at least one person at Tier 3.
• AI development or high-risk deployment (proprietary models, AI in HR decisions, credit scoring): Tier 1 for everyone, Tier 2 for specialist departments, Tier 3 as a core competency within the team.

Practical example: midsize mechanical engineering company with 120 employees

A mechanical engineering company from the Stuttgart region with 120 employees already uses AI in three areas: predictive maintenance in manufacturing, AI-powered quotation calculations in sales, and an internal RAG system for technical documentation.

The company implemented the IHK competency model as follows:

• Tier 1 (Foundation): 95 employees—two-day foundational training. Cost: 95 x €800 = €76,000
• Tier 2 (Practitioner): 20 employees from sales, service, and manufacturing. Cost: 20 x €3,500 = €70,000
• Tier 3 (Expert): 5 employees from IT and production management. Cost: 5 x €6,000 = €30,000
• Total investment: €176,000, spread over 12 months
• Average per employee: approximately €1,470

Against this stand estimated efficiency gains of €280,000 per year—through faster quotation preparation (minus 35 percent processing time), reduced unplanned machine downtime (minus 22 percent), and better first-resolution rates in service. The return on investment was thus achieved in under eight months.

Practical guide—building an AI training program in 5 steps

The theory is clear, but how do you concretely implement an AI training program? The following guide describes five steps that have proven effective in practice.

Step 1: assessment (weeks 1 to 2)

Before you train, you need to know where you stand. Conduct a systematic assessment:

• Create an AI inventory: Which AI systems are being used in the company? Informally used tools such as ChatGPT or DeepL count as well.
• Competency mapping: Where do your employees stand? An anonymous self-assessment combined with short knowledge tests provides a realistic picture.
• Risk analysis: Which of your AI applications falls under which risk class of the EU AI Act?

Step 2: define the training architecture (weeks 3 to 4)

Based on the assessment, determine who needs to reach which competency tier:

• Assign each employee to one of the three tiers.
• Define learning objectives per tier that align with your specific AI applications.
• Choose training formats: in-person seminars, e-learning, blended learning, or innovative approaches such as VR training.

In this context, a pilot project by Lufthansa delivers remarkable results: The company is testing VR-based AI training for employees. The learning speed is 40 percent higher than that of traditional training, and the transfer rate—meaning the proportion of what was learned that is actually applied in daily work—reaches 78 percent. With traditional in-person training, this figure is only 23 percent. While VR training may not be feasible for every company, the example demonstrates how important the choice of the right format is.

Step 3: identify internal champions (weeks 3 to 4)

Every successful training program needs internal multipliers:

• Identify one to two tech-savvy employees in each department who can serve as “AI champions.”
• Train these individuals first at Tier 2 or 3.
• AI champions become the first point of contact for colleagues and bridge the gap between theory and practice.

This approach is particularly effective in addressing the identified generational gap: When the experienced 55-year-old design engineer learns from his same-age colleague in the neighboring office how a RAG system accelerates technical documentation search, that is more convincing than any external seminar.

Step 4: conduct and document training (weeks 5 to 12)

Three aspects are critical during implementation:

• Practical relevance: Every training unit should be linked to a concrete task from the participants’ daily work. Abstract theory without practical application fades quickly.
• Documentation: Record for each participant: date, topic, scope, competency tier, and result. This documentation is your proof for the supervisory authority.
• Address anxiety: Take the fear of job loss seriously. Communicate clearly that AI training does not mean replacing jobs but enabling employees to become more productive and valuable.

Step 5: continuous development (from week 13 onward)

An AI training program is never “finished”:

• Schedule quarterly refreshers that cover new tools and regulatory changes.
• Establish an internal knowledge platform where employees can share experiences and best practices.
• Measure success: Not just participation numbers, but actual application in daily work, time savings, and quality improvements.
• Adjust the training architecture annually to meet evolving requirements.

Frequently asked questions

Does Article 4 also apply to sole proprietors and freelancers?

Yes. Article 4 of the EU AI Act does not distinguish by company size. Its wording targets staff and persons who work with AI systems on the company’s behalf, such as freelancers. For sole proprietors, recording their own training still makes sense. The measures depend on context: a freelance copywriter who uses ChatGPT does not need a data science certificate.

What specific training documentation does the authority require?

The EU AI Act does not prescribe a specific certificate or format. What matters is that you can show which measures you have taken. In practice, a combination of training records (date, participants, content, duration), learning checks, and a regularly updated competency matrix works well. A certificate from a recognized provider can make proof easier but is not a requirement.

What happens if an employee refuses to participate in training?

Responsibility lies with the company, not with the individual employee: Article 4 obliges providers and deployers. Under employment law, employers can make participation in AI training a work instruction, provided it takes place during working hours and is paid. It is advisable, however, to rely on persuasion rather than compulsion; experience shows that practical training with immediately noticeable benefits achieves the highest acceptance.

Is a one-time training session enough to comply with Article 4?

Hardly. The obligation applies as long as AI systems are in use, and the measures must fit the context of use. Because AI technologies and the regulatory framework keep evolving, training programs should be updated regularly. The IHK network recommends refresher training at least once a year, plus training whenever new AI systems are introduced or existing systems receive significant updates.

What are the realistic costs for a midsize company with 20 employees?

Based on the IHK recommendations, you can expect the following costs: Tier 1 foundation training for all 20 employees (20 x €800 = €16,000), Tier 2 training for 3 to 5 power users (4 x €3,000 = €12,000), and optionally Tier 3 training for one AI lead (1 x €5,000 = €5,000). Total cost: €28,000 to €33,000 in the first year, significantly less in subsequent years due to refresher-only courses. Many IHK members can apply for subsidies or grants for continuing education measures that partially offset these costs.

References

• EU AI Act—Article 4 (AI Literacy): Official Artificial Intelligence Regulation of the European Union. Full legal text and explanations at https://artificialintelligenceact.eu/de/article/4/
• Fraunhofer Academy—AI Literacy Study Germany 2026: Comprehensive survey on the state of AI literacy in German companies, published in March 2026. Details at https://www.academy.fraunhofer.de/
• Bitkom—Continuing Education Survey 2026: Representative survey among German companies on continuing education priorities and AI training measures. Results at https://www.bitkom.org/
• IHK Network—Position Paper on AI Competency Model: Recommendations for a tiered competency model for implementing the AI literacy obligation under Article 4 of the EU AI Act. Available at https://www.ihk.de/
• Handelsblatt—VR-Based AI Training at Lufthansa: Report on the pilot project with virtual reality training formats and their results compared to traditional in-person training. Published on March 5, 2026 at https://www.handelsblatt.com/