AI Act Training: proving AI literacy under Art. 4 (funding and deadline 2026)
AI literacy under AI Act Art. 4: in force since 02.02.2025, enforcement from 02.08.2026. What Art. 4 requires, what the proof of competence looks like, who is liable, and how QCG funding works.
Most German Mittelstand companies do not know they have to train their staff in the use of AI, and even fewer can prove that training. But that is exactly what the EU AI Act requires in Article 4. The AI literacy obligation has applied since 02.02.2025. What arrives on 02.08.2026 is enforcement. An AI Act training worth its name does not just convey content, it produces the per-person proof an auditor wants to see.
I (Sebastian) see the same gap in our workshops almost every week: ChatGPT, Copilot and Claude are long in use, often in shadow-IT mode, but there is no documented training and no proof of competence. That is not a tool problem, it is a compliance and liability problem.
What Article 4 actually requires
Article 4 of the EU AI Regulation (Regulation 2024/1689) obliges providers and deployers of AI systems to ensure a sufficient level of AI literacy among their staff. A Mittelstand company that uses ChatGPT or Copilot is a deployer. The obligation has been in force since 02.02.2025. General enforcement starts on 02.08.2026, alongside the full applicability of most AI Act obligations.
Important for context: Article 4 itself has no direct fine catalogue in the AI Act. Article 99(1) delegates sanctions to the member states. In Germany, as of May 2026, no specific Art. 4 fine schedule has been published. That does not make Art. 4 toothless: expect administrative proceedings with warnings and remediation deadlines first, and depending on severity, managing-director liability comes into play. Anyone quoting 35 million euros or 7.5 million for Art. 4 is confusing it with Art. 5 (prohibited practices) or Art. 99(5) (false information).
The proof of competence: what an auditor wants to see
Competence is not the same as attendance. The proof that counts in an audit is per-person and documents four things for each employee: who was trained, when, on what content and with what test result. Without this documentation, the default in case of doubt is: not trained.
In practice that means an AI Act training has to output a PDF per person documenting these four points, plus a central overview of the trained workforce. This is exactly where awareness videos and one-off town halls fail: they may convey content, but they produce no auditable proof.
The 3 training tiers
AI literacy is role-dependent. A sensible AI Act training staggers into three tiers instead of showing everyone the same thing.
Tier 1, baseline literacy for the whole workforce. What is generative AI, what may go in and what may not (data protection, trade secrets), where are the limits (hallucination, no legal advice). This is the baseline Art. 4 requires for everyone affected.
Tier 2, role literacy for active users. For teams that use AI day to day: safe prompting, tool-specific data policy, handling output quality, escalation under uncertainty. This is where the productivity lever sits.
Tier 3, deep literacy for developers and AI owners. For those who build or own AI: high-risk classification under Annex III, human oversight under Art. 14, eval governance, vendor selection. This overlaps with our coding-agent trainings.
Funding through the Qualifizierungschancengesetz
An AI Act training can be funded through the German Qualifizierungschancengesetz (Section 82 SGB III) when it teaches skills that go beyond short-term, purely job-specific adjustment. The funding rate for course costs is staggered by company size, with thresholds at 50 and 500 employees: for companies with fewer than 50 employees, an employer cost contribution can be waived, so up to 100 percent of course costs are funded. For 50 to 499 employees, up to 50 percent; for 500 and more, up to 25 percent. The wage subsidy is staggered at 75, 50 and 25 percent.
One important limit that many consultant tables omit: the funded measure must exceed 120 hours. A short one-to-two-hour awareness session therefore falls outside QCG funding. It becomes eligible when the AI Act training is part of a larger qualification programme. The full overview is in AI funding for the Mittelstand 2026. The numbers come from Section 82 SGB III, not from consultant tables (the circulating 10/250/2,500 staggering is wrong). (As of May 2026)
In-house or open course
For baseline literacy (Tier 1), an open course is often the fastest solution, especially for distributed teams. You can find our open formats under our courses. For Tier 2 and 3, where your tools, your data policy and your proof matter, an in-house training pays off. How we set up in-house projects is under how we work.
FAQ
Does Art. 4 apply to us if we only use AI, not develop it? Yes. The obligation hits deployers, not just providers. Anyone using ChatGPT, Copilot or Claude in the company is a deployer and must ensure AI literacy.
Is a one-off awareness video enough? Hardly, for the proof. It lacks the per-person test result and the documentation per employee. An auditor wants to see who learned and passed what, and when.
What penalty actually applies? Art. 4 has no direct AI Act fine catalogue (Art. 99(1) delegates to member states). Realistically, administrative proceedings with warnings come first, repeat cases bring fines and possible managing-director liability. The 35-million figure belongs to Art. 5, not Art. 4.
Is the training eligible for funding? Through the QCG (Section 82 SGB III) yes, if the measure exceeds 120 hours and goes beyond pure job adjustment. A short awareness unit alone is not enough. (As of May 2026)
Read more
- AI literacy mandate from August 2026: executive checklist
- AI Act Art. 50: transparency obligation from August 2026
- EU AI Act fines myth: what actually applies
- AI Act 90-day compliance plan for engineering teams
- AI funding for the Mittelstand 2026
- GitHub Copilot training for engineering teams
- In-house training vs. open AI courses in the Mittelstand
Sources
- EU AI Act, Regulation (EU) 2024/1689, Art. 4 (AI literacy), Art. 99 (sanctions, esp. 99(1) delegation to member states), Art. 113 (timeline), eur-lex.europa.eu
- Section 82 SGB III, https://www.gesetze-im-internet.de/sgb_3/__82.html; Federal Employment Agency, administrative directive on Section 82 SGB III, effective 01.01.2026
- Sentient Dynamics workshop aggregate (DACH Mittelstand clients, 2025-2026)
How Sentient Dynamics can help
We set up AI Act trainings that ship the proof with them: three tiers by role, a per-person proof-of-competence PDF, a central workforce overview and, where it makes sense, embedded in a QCG-eligible programme.
About the author
Co-Founder · Business & Content Lead
Co-Founder von Sentient Dynamics. 15+ Jahre Business-Strategie (u.a. SAP), MBA. Schreibt über AI-Act-Compliance, ROI-Messung und wie Mittelstand-CTOs agentische KI tatsächlich einführen.