How a city administration made its workforce AI-ready (Art. 4 EU AI Act)

A medium-sized German city administration wanted to use artificial intelligence in everyday work and at the same time had to enable its workforce, in a legally compliant way, to acquire AI literacy under Article 4 of the EU AI Act. The challenge lay less in the legal text than in acceptance: Some employees viewed AI with skepticism. This case study shows how AI training in public administration succeeds when compliance with obligations and trust-building come together – and how the competence acquired can be proven.

Starting point: the AI obligation meets apprehension

The task was twofold. On the one hand, regulation: Since February 2, 2025, Article 4 of the EU AI Act (Regulation (EU) 2024/1689) has required every organization that uses AI systems to take measures to support the development of the "AI literacy" of its staff (according to the version under Regulation (EU) 2026/1744 applicable since July 27, 2026; previously: measures to ensure, to their best extent, a sufficient level of AI literacy). What this means is outlined in Article 3 no. 56 of the Regulation: the skills, knowledge, and understanding that allow an informed deployment of AI systems and awareness of their opportunities and risks. This obligation has no sectoral exemption for the public sector – public authorities and municipalities are addressed just as much as companies. The Regulation's general date of application was August 2, 2026; its chapter on penalties, except Article 101, has applied since August 2, 2025 (Art. 113 EU AI Act). For a municipality in Germany, the authority competent under state law has been the market surveillance authority since July 29, 2026 (Section 2 (6) KI-MIG). If it finds a violation, it notifies the authority responsible for legal or technical supervision and, as a rule, gives it the opportunity to comment before taking measures of its own (Section 9 (3) KI-MIG). Fines are not imposed on authorities and other public bodies within the meaning of Section 2 (1) or (2) BDSG, which include municipalities, in the cases of Section 15 (1) and (2) KI-MIG and Art. 99 (3) to (5) EU AI Act (Section 17 (2) KI-MIG); a violation of Article 4 is not included in these catalogs of fines in any case. The project described still ran under the original version of Article 4.

On the other hand, the people: For some of the workforce, especially the more experienced age groups, dealing with AI was genuinely new territory and associated with noticeable uncertainty. In everyday administrative work, this involved practical areas of application – such as support with drafting texts, preparing information, or answering recurring inquiries. Purely factual mandatory training on the Regulation would hardly have dispelled these reservations and might even have reinforced them. The administration was therefore looking for professional development that would achieve both: meet the legal requirements and meet the workforce where they are with their specific questions and concerns.

The challenge: fulfilling the obligation and creating acceptance

The crux lay not in the legal content but in how it was conveyed. The workforce was heterogeneous – from AI enthusiasts to employees who had never had any contact with such tools. There was also the justified concern about data protection, because in everyday administrative work sensitive personal data can quickly end up in an AI system if the boundaries are unclear. Public authorities in particular process large volumes of sensitive data belonging to citizens, so careless handling of AI tools can quickly become problematic under data protection law. And at the end, there was to be reliable proof of AI literacy without the training being perceived as a mere compulsory exercise. The human level therefore had to be taken just as seriously as the regulatory level – otherwise the obligation would have been met formally but would have remained ineffective in everyday work.

The solution: regulation and empathy in one professional development course

A note on our own behalf: Bridgly implemented intensive AI professional development that deliberately goes beyond mere compliance with obligations. Unlike with a self-built learning offering, the administration used a fully rolled-out training course including a learning platform – content, process, and record-keeping were coordinated from the outset. In terms of content, the professional development conveyed the requirements of the EU AI Act and translated them into concrete use cases from everyday administrative work. At least as important was the second strand: The training took fears and reservations seriously and actively worked through them instead of ignoring them.

Instead of imposing technical terms, the professional development built trust step by step. It made clear what AI can do, what it cannot do, and where its ethical limits lie. This approach was the decisive difference from classic mandatory training – and the reason why the workforce embraced the content.

Implementation step by step

Making the regulatory basics tangible

The starting point was the requirements of Article 4 EU AI Act – in understandable language instead of legal overload, so that it becomes clear why the training takes place and what it protects. In this way, an abstract legal obligation became a comprehensible reason to engage with one's own AI tools.

Administration-specific use cases

Generic tech examples seem pale in the context of public authorities. The professional development therefore worked with situations from everyday administrative work and used them to make both the opportunities and the limits of AI use concrete – including the question of which data belongs in an AI system and which does not.

Taking fears seriously, building trust

A separate strand was dedicated to the concerns of the workforce. Through transparent explanation, diffuse skepticism turned into an informed attitude: If you know the limits of the technology, you do not have to fear it and can use it with confidence.

Documenting AI literacy traceably

So that training participation can also be proven, the training ran via a learning platform that logs assignment, completion, and date for each person. This creates documented proof of training participation – retrievable should a supervisory authority want to inspect it in the future. The EU AI Act does not prescribe proof, or a format for proof, for Article 4; what matters is that participation can be demonstrated in a traceable way when needed.

The result: from uncertainty to practical competence

In this administration, AI has gone from an anxiety-laden topic to a comprehensible tool in everyday work. The workforce now has a well-founded understanding of AI that provides security both professionally and emotionally: The regulatory basics have been conveyed, and uncertainty has turned into genuine practical competence. Equally important for the administration: Participation is documented and can therefore be evidenced at any time; Article 4 does not require formal proof, but the Federal Network Agency recommends documenting the measures. What was decisive was that the professional development reached both the professional requirements and the emotional level of the workforce – thus turning mere compliance with obligations into real acceptance.

AI training in public administration: what others can take away

• Think obligation and acceptance together. AI training that merely ticks off the Regulation does not change attitudes. Only the combination of regulation and reservations that are taken seriously creates competence that is effective in everyday work.
• Start with everyday administrative work. Examples close to public authorities have a stronger effect than abstract technology topics – they show directly where AI helps and where caution is needed.
• Think about proof from the outset. If you document the measures for AI literacy under Article 4 via a learning platform, you are prepared for later audits.
• Rely on ready-made solutions. A rolled-out training course including record-keeping relieves scarce administrative resources more than a self-assembled offering.

We collect further practical examples from companies and organizations in our case studies – including how an AI start-up set up its AI training under Article 4.

Do public authorities have to train their employees in AI?

In a broader sense, yes: the provision does not prescribe a particular training format. Since February 2, 2025, Article 4 of the EU AI Act has obliged all organizations that use AI systems to take measures to support the development of the AI literacy of their staff (version since July 27, 2026). This obligation applies regardless of legal form and therefore also to public authorities, municipalities, and public institutions.

How do you deal with AI skepticism among the workforce?

What works is not ignoring reservations but actively addressing them. When training makes transparent what AI can do, where its limits lie, and how data should be handled, diffuse skepticism turns into an informed attitude that enables people to act.

Is e-learning sufficient for AI professional development in public administration?

E-learning is well suited for conveying the basics and for documented proof because it works independently of location and time. What matters is that the content is current, close to administrative practice, and presented in a comprehensible way.

How is AI literacy proven?

Through documented participation with traceable knowledge transfer. A learning platform that logs assignment, completion, and date for each person makes participation evidenced at any time. Article 4 EU AI Act does not prescribe formal proof; the Federal Network Agency recommends documenting the measures.

• Regulation (EU) 2024/1689 (EU AI Act), Art. 4 – AI literacy, EUR-Lex – accessed 14 Oct 2026.
• Regulation (EU) 2024/1689 (EU AI Act), Art. 3 no. 56 – definition of "AI literacy", EUR-Lex – accessed 14 Oct 2026.
• Regulation (EU) 2026/1744 ("Digital Omnibus"), Art. 1 no. 5 – new version of Art. 4 EU AI Act, in force since 27 Jul 2026: EUR-Lex – accessed 14 Oct 2026.
• Regulation (EU) 2024/1689 (EU AI Act), Art. 99 (penalties) and Art. 113 (entry into application): EUR-Lex – accessed 14 Oct 2026.
• KI-Marktüberwachungs-und-Innovationsförderungs-Gesetz (KI-MIG), BGBl. 2026 I no. 223, in force since 29 Jul 2026, Section 2 (6), Section 9 (3), Section 15, and Section 17 (2): gesetze-im-internet.de – accessed 14 Oct 2026.
• Section 2 Federal Data Protection Act (BDSG) (public bodies of the Federation and the federal states): gesetze-im-internet.de – accessed 14 Oct 2026.
• Federal Network Agency, AI literacy (guidance on documenting the measures): bundesnetzagentur.de – accessed 14 Oct 2026.
• Featured image: AI-generated; labeled "Created with AI" as an overlay in line with Art. 50 EU AI Act. Proof of license/creation will be added during image production (provider/date).