AI literacy (Article 4)
Structured, documented training with a certificate - the obvious measure Article 4 asks for, and easy to evidence. AZAV-compliant and fundable up to 100% via § 82 SGB III.
Go to AI ACADEMY →AI COMPLIANCE · KI. ABER SICHER.
EU AI Act. Pragmatic, not panicked.
The EU AI Act applies - but not every obligation applies to every company. We first determine which risk class your AI systems fall into and which obligations actually follow. Then we implement exactly that: AI literacy, AI policy, governance. No more, no less.
Which obligations apply since when?
| Since | What applies |
|---|---|
| 02 Feb 2025 | Prohibited AI practices (e.g. social scoring) and Article 4: companies using AI must take measures to support the AI literacy of their staff. |
| 02 Aug 2025 | Obligations for providers of general-purpose AI models (GPAI), governance and penalty rules. |
| 27 Jul 2026 | Digital Omnibus (Regulation (EU) 2026/1744) in force: Article 4 now requires measures instead of a specific literacy level, and the high-risk deadlines are postponed. |
| 02 Aug 2026 | General applicability: transparency obligations under Article 50 (label AI chatbots, disclose deepfakes). |
| 02 Dec 2026 | End of the transition period for machine-readable marking of AI-generated content (Article 50(2)) for systems already on the market before 02 Aug 2026. |
| 02 Dec 2027 | Obligations for high-risk systems under Annex III (e.g. AI in recruiting or credit scoring). |
| 02 Aug 2028 | High-risk obligations for AI in regulated products (Article 6(1)). |
As of 10/2026, after the Digital Omnibus (Regulation (EU) 2026/1744). Factual summary - not legal advice.
Quick check: does the labelling duty apply to you?
Five questions, instant result - no sign-up.
- Do you run a chatbot, voice or phone assistant that customers, applicants or partners write or talk to directly?
- Do you publish AI-generated or AI-altered images, videos or voices that look real - such as real people, places or events?
- Do you publish AI-generated text informing the public on matters of public interest without a human reviewing it and taking editorial responsibility?
- Do you use AI to recognise people’s emotions or to assign people to categories based on biometric data?
- Do you offer an AI system under your own name - for example your own bot for customers or a generator you built yourself?
The four risk classes
The EU AI Act regulates by risk. The first step of any compliance work is therefore classification: which of your AI systems fall into which class?
| Risk class | Examples | Consequence |
|---|---|---|
| Unacceptable risk | Social scoring, manipulative systems | Prohibited (since 02/2025) |
| High risk | AI in recruiting, credit scoring, critical infrastructure | Extensive obligations from 02 Dec 2027: risk management, documentation, human oversight |
| Limited risk | Chatbots, generative AI, deepfakes | Labelling duty under Article 50 (since 02 Aug 2026) |
| Minimal risk | Spam filters, simple assistance systems | No specific obligations - Article 4 (AI literacy) still applies |
We classify your AI systems and derive the concrete obligations - as part of the AI CHECK or as standalone compliance consulting.
Our compliance services
Compliance package
Training plus AI policy plus data protection impact assessment (DPIA): the organizational baseline for using AI in your company.
Check funding →AI governance & ISO 42001
Governance structures, risk analysis, and documented processes - aligned with the international standard for AI management systems (ISO/IEC 42001).
Start with the AI CHECK →Compliance AI technology
AI agents that audit other AI systems - our award-winning technology approach for continuous control instead of one-off audits.
View AI EXPERT →Frequently asked questions
Does the EU AI Act apply to small and mid-sized companies?
Yes. The regulation has no general SME exemption. Article 4 (AI literacy) has applied since 02 Feb 2025 to every company using AI - regardless of size. The scope of the remaining obligations depends on the risk class of the systems used, not on company size.
What are the penalties for violations?
Fines of up to €35 million or 7% of global annual turnover for prohibited practices; up to €15 million or 3% for other violations. In addition, there is liability risk if literacy and documentation obligations cannot be demonstrated in case of damage.
Is internal training sufficient proof of AI literacy?
Yes, if you can evidence it. Article 4 prescribes no curriculum and, since the Digital Omnibus (27 Jul 2026), requires measures to support AI literacy rather than a specific level. A structured, documented training with a certificate is the simplest evidence. Our AZAV-compliant programs deliver exactly that - partly up to 100% government-funded.
Which AI systems does the labelling duty cover?
Article 50 covers four groups: systems interacting directly with people (chatbots, voice assistants, AI telephony), systems generating synthetic audio, image, video or text, emotion recognition and biometric categorisation, and deepfakes. It does not cover purely internal AI analytics with no outside contact. Your role also decides: in the first two cases the provider of the system is liable, in the last two the deployer - you, as the company using it.
Do we have to label our website chatbot as AI?
Yes, unless it is obvious from the context that this is an AI - and the yardstick is stricter than many assume: a window merely labelled "Chat" does not clear it. One sentence when the chat window opens usually suffices. Timing matters: under Article 50(5) the notice must be visible at the first interaction, so an entry in the imprint or terms of use does not qualify.
Do we have to label AI-written text?
As a rule, no. The disclosure duty for text only applies to publications informing the public on matters of public interest - and even there it falls away if the text was reviewed by a human and a natural or legal person holds editorial responsibility. Product descriptions, proposals and professional articles with a named responsible party therefore regularly fall outside it.
What is a high-risk system?
AI systems in the areas listed in Annex III - such as recruiting, credit scoring, critical infrastructure, or access to education - as well as AI as a safety component of regulated products. They carry the most extensive obligations: risk management, data quality, documentation, human oversight. After the Digital Omnibus these apply from 02 Dec 2027 (Annex III) and 02 Aug 2028 (regulated products).
Does your consulting replace legal advice?
No. We advise on the technical-organizational side: risk classification, literacy building, policies, governance processes. For binding legal questions we work with your law firm or recommend specialized partners.
Article 50 in detail: the four cases of the labelling duty →
Where do you stand on the EU AI Act?
In a free initial consultation we clarify which of your AI systems are affected and which obligations follow - and whether training and consulting are fundable for you.