Since 2 February 2025, Article 4 of the EU AI Regulation has applied. It obliges providers and operators of AI systems to ensure that their staff has a sufficient level of AI competence. Anyone with just ChatGPT open in the browser is an operator. The obligation therefore affects far more companies than expected, and it applies now, not from 2026.
What Article 4 requires
The wording is short. Providers and operators of AI systems take measures to ensure, to the best of their efforts, that their staff and other persons operating or using AI systems on their behalf have a sufficient level of AI competence, taking into account technical knowledge, experience, training and the context of use. The regulation prescribes no certificates and names no number of hours. It requires companies to assess and be able to demonstrate the competence of their people.
- Who
- Providers and operators of AI systems. An operator is any company whose staff use an AI system on its behalf, including a simple chatbot subscription.
- What
- AI competence of staff and contracted third parties, to the best of efforts, appropriate to the context of use.
- Since when
- 2 February 2025. Chapter I of the Regulation entered into force on that day; Article 4 is part of it.
- Proof
- The regulation names no format. A documented training with attendance list and contents is common and auditable.
- Risk
- Violations of Article 4 can be sanctioned as violations of obligations under the regulation. The fine framework for such violations is up to 15 million euros or 3 percent of global annual turnover, whichever is higher.
Who the obligation really affects
The obligation does not depend on whether a company develops AI. It depends on staff using AI systems. A mid-sized company whose sales team writes texts with ChatGPT is an operator. A law firm doing research with an AI tool is too. The scope of the competence obligation differs: operating a high-risk system requires deeper knowledge than a team that occasionally generates image ideas.
In practice this means: an overview of all AI tools in the company is the first step. The next question is which employees work with them and what they need to know.
What an AI training should cover
Because the regulation does not prescribe content, companies orient themselves on what auditors and insurers expect: traceable content, documented participation, reference to daily work. This framework has proved itself.
- Overview: what language, image and video models can do, and where reliability ends.
- Daily work: prompts for own tasks, on real examples from the company.
- Limits: where models make mistakes, where facts are missing, where human review remains mandatory.
- Rules: data protection, confidentiality, labelling obligations for AI content.
- Evidence: attendance list, content, date. That is the training file for Article 4.
Subsidies: what actually applies
Many providers advertise subsidisability. The longest federal funding line, further training for employed workers under § 82 SGB III, has clear limits: the measure must last more than 120 hours, and the provider must be approved. A half day or a single workshop day falls outside. Subsidies depending on the federal state or via education vouchers for individual employees remain case by case. If you need funding, clarify with your chamber of commerce or the Federal Employment Agency what applies in your case.
What we recommend
Start small and document. A half-day overview for the whole team, then a practice day for the people who work with the tools daily, and a module that fills the training file for Article 4. QLTY media offers these three formats in Cologne, in-house and online, at fixed prices. Formats and prices are listed here.







AI Labelling Obligations: What Has Applied Since 2 August 2026
AI in Marketing: The Current Figures