AI Literacy Training Curriculum (Art. 4): 8-Module Course for SMEs

Obligation and recommendation: in this article, an obligation is whatever is formulated as an obligation with a source reference. Time figures, numbers of questions, pass marks, retention periods and repetition intervals are practical values from the first version and are expressly marked as practitioner recommendations. Market and tool information remains unchanged at the status of the first version of 2 May 2026.

1. Art. 4 EU AI Act: obligation, concept, timeline

Art. 4 EU AI Act requires providers and deployers of AI systems to take measures to support the development of AI literacy of their staff — the obligation has applied since 2 February 2025 and has therefore already taken effect; it is not a future one. Its current wording stems from Regulation (EU) 2026/1744, in force since 27 July 2026.

"Providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf... This obligation does not require providers or deployers to guarantee any particular level of AI literacy for any person." — Art. 4 AI Act in the version applicable since 27 July 2026

The legal term is 'AI literacy' and it is defined in Art. 3 no. 56 AI Act: the skills, knowledge and understanding that allow AI systems to be deployed in an informed way and that create awareness of the opportunities, risks and possible harm. The yardstick is therefore a measures-based one: under the wording applicable since 27 July 2026, measures must be taken that support the development of AI literacy while taking into account technical knowledge, experience, education and training as well as the context of use; a specific level of competence is expressly not owed. The provision prescribes neither a number of hours, nor a number of questions, nor a certificate format — according to the Commission FAQ (non-binding), no certificate is required and an internal training record suffices; Art. 4(2) and (3) add support measures from the Commission and the Member States as well as recommendations from the AI Board. That is precisely why all the figures in the following sections are practical values and not legal obligations.

Placing the obligation in the overall timeline of the AI Act matters for training planning, because Modules 5 to 7 deal with obligations that became applicable at different points in time. All dates in the following table are dates of application under the AI Act — not practitioner recommendations and not internal deadlines.

Art. 4 has been recast by Regulation (EU) 2026/1744 (Digital Omnibus AI), in force since 27 July 2026, but its start of application has not been postponed — the training obligation has applied unchanged since 2 February 2025 and does not depend on the high-risk deadlines.

2. Who the obligation applies to: providers, deployers, contractors

Art. 4 addresses both roles under the AI Act — providers and deployers — and, beyond your own staff, expressly covers other persons dealing with the operation and use of AI systems on your behalf. For course planning this means: the distribution list is not the payroll, but the group of those who operate AI systems or use their outputs.

A provider is a party that develops an AI system or places it on the market; a deployer is a party that uses an AI system under its own authority Art. 3 no. 3 and no. 4 AI Act. The role is not fixed permanently: where the purpose or functionality is substantially modified, or where the system is placed on the market under its own name, a deployer moves into the provider role Art. 25 AI Act. A RAG system built on a GPAI interface and marketed as a standalone assistant is the standard example of this.

The question of role does not determine whether training takes place — Art. 4 applies to both roles — but how deep Module 5 and Module 7 need to go: for high-risk AI systems, providers need risk management (Art. 9), technical documentation under Annex IV (Art. 11), conformity assessment (Art. 43) and registration in the EU database (Art. 49) as training content, while deployers need the relevance check for input data (Art. 26(4)), the retention of logs for at least six months (Art. 26(6)), the information of workers (Art. 26(7)) and — for public bodies or Annex III no. 5(b)/(c) — the FRIA (Art. 27).

Practitioner recommendation: build the training distribution list from the AI system inventory, not from the staff list. Six fields are recorded per system — name and provider, use case, role, risk class, types of data processed including special categories, and those responsible for oversight and escalation. Anyone entered there as exercising oversight or as a user group belongs on the distribution list.

3. 5 target group personas

Five personas cover the typical SME workforce, because AI obligations differ according to decision-making responsibility, system access and type of data — not according to hierarchy. The hours given in the table are practical values from the first version for the role-specific in-depth training that comes on top of the core curriculum from section 4 (approx. 2.75 hours); Art. 4 AI Act in the version applicable since 27 July 2026 requires measures that support the development of AI literacy and prescribes neither a number of hours nor a form of record — according to the Commission FAQ (non-binding), no certificate is required and an internal training record suffices.

Which modules are covered in depth per persona follows directly from this: management needs Module 2 and the penalty tiers from section 9, plus — only for entities covered by NIS2 — the management body's own training and supervision obligation (section 7; whether your entity is covered is explained in section 8). IT, compliance and data protection officers maintain the inventory and the classification and need Module 5 in full depth. HR is the persona with the highest double burden: assessment systems in recruiting fall under Annex III no. 4, and discrimination liability applies irrespective of the AI Act deadlines § 22 AGG § 12(12), § 26(12) GlBG. Marketing and sales generate synthetic content and run chatbots — for them, Module 6 is the core module. For all remaining employees, Module 3 (prohibitions) and Module 4 (day-to-day work, data protection, output quality control) are the priority.

4. 8-module curriculum

The curriculum consists of eight modules that build on one another from the definition of AI through to the final quiz: Modules 1 to 3 establish the concept, the regulatory logic and the prohibitions, Modules 4 to 7 translate them into day-to-day work and into high-risk, transparency and GPAI obligations, and Module 8 tests. The minute figures are practical values from the first version.

1. Module 1: what is AI? (15 min) — definition in Art. 3, machine learning, GPAI. What is taught is the legal definition: a machine-based system designed to operate with varying levels of autonomy, that may exhibit adaptiveness after deployment and that infers, from the input it receives, how to generate outputs such as predictions, content, recommendations or decisions Art. 3 no. 1 AI Act. Alongside this, the distinction from a general-purpose AI model, which displays significant generality and can be integrated into a variety of downstream systems Art. 3 no. 63 AI Act. Why this comes first: without this distinction, the inventory ends up containing either too many or too few systems.
2. Module 2: EU AI Act overview (20 min) — risk-based approach, obligations of providers/deployers. The core is the regulatory logic: the type and content of the rules are tailored to the intensity and scope of the risk, which gives rise to prohibited practices, requirements for high-risk AI systems and transparency obligations for certain AI systems recital 26 AI Act. Added to this are the allocation of roles from section 2 and the timeline from section 1.
3. Module 3: prohibited practices under Art. 5 (15 min) — social scoring, emotion detection, manipulative AI. Art. 5(1) prohibits eight categories, including subliminal or purposefully manipulative techniques causing significant harm ((a)), the exploitation of vulnerabilities ((b)), social scoring ((c)), predictive assessment of criminal offences based solely on profiling ((d)), untargeted scraping of facial images ((e)), emotion recognition in the workplace and in education institutions ((f)), biometric categorisation to infer sensitive attributes ((g)) and real-time remote biometric identification in publicly accessible spaces for law enforcement purposes ((h)) Art. 5(1) AI Act. Why this module concerns all personas: (f) hits SMEs directly as soon as tools offer sentiment or attention analysis in the employment context — and this infringement carries the highest range of fines in the Regulation (section 9).
4. Module 4: AI in day-to-day work — what am I allowed to do? (30 min) — AUP, data protection, confidentiality, verification. The longest module, because this is where the acceptable use policy meets concrete inputs: every processing of personal data needs a legal basis Art. 6(1) GDPR, special categories such as health, biometric or trade union data are subject to a general prohibition on processing with narrow exceptions Art. 9(1) GDPR, and using a tool that processes on your behalf requires a data processing agreement with the statutory minimum content Art. 28(1) and (3) GDPR. Verification means: outputs are checked before use, because responsibility for the result remains with the deployer.
5. Module 5: high-risk AI (20 min) — Annex III, FRIA, conformity assessment. A system is high-risk if it is intended to be used as a safety component of a product covered by Annex I, or is itself such a product and is subject to third-party conformity assessment, or if it falls under Annex III Art. 6(1) and (2) AI Act. Annex III names eight areas: biometrics, critical infrastructure, education and vocational training, employment and worker management, access to essential services including creditworthiness and health/life insurance, law enforcement, migration/asylum/border control, and administration of justice and democratic processes Annex III AI Act. The exception in Art. 6(3) is to be interpreted narrowly; if the system generates recommendations or assessments that can materially influence human decisions, it remains high-risk AI.
6. Module 6: transparency under Art. 50 (15 min) — chatbot notices, deepfake labelling, watermarking. Four obligations across two roles: providers must design AI systems intended to interact directly with people so that the person is informed about the AI interaction (paragraph 1), and must mark synthetic audio, image, video or text outputs in a machine-readable format as artificially generated or manipulated (paragraph 2); deployers must inform the persons concerned in cases of emotion recognition and biometric categorisation (paragraph 3) and disclose deepfakes (paragraph 4) Art. 50(1) to (4) AI Act. These obligations have been applicable since 2 August 2026.
7. Module 7: GPAI + ChatGPT/Claude/Copilot (20 min) — acceptable use, data flow, loss of creativity. The most important message for SMEs: anyone who merely uses GPAI models is a deployer, not a provider. Provider obligations arise only where you train your own model, make substantial fine-tuning modifications or engage in white-label marketing; they comprise technical documentation under Annex XI, information for downstream providers, a copyright strategy and a public summary of the training content Art. 53(1) AI Act. Models with systemic risk — presumed from a cumulative training compute of more than 10^25 floating point operations — additionally carry model evaluation including adversarial testing, risk mitigation, incident reporting and cybersecurity obligations Art. 51(2), Art. 55(1) AI Act.
8. Module 8: knowledge quiz + certificate (30 min) — 25 questions, 70 % pass mark, certificate generation. The structure and the pass mark are practitioner recommendations from the first version; details in section 5.

On the arithmetic of time: the eight module times add up to 165 minutes, i.e. approx. 2.75 hours of core curriculum — this is the reference figure for the course. The hours in the personas table in section 3 come on top as role-specific in-depth training. Both are practical values from the first version, not requirements; what matters under Art. 4 AI Act in the version applicable since 27 July 2026 are the measures taken to support AI literacy — a specific level of competence is not owed.

5. Structure of the knowledge quiz

In the first version, the final quiz tests 25 questions with a pass mark of at least 70 % correct answers — both practitioner recommendations from the first version, not requirements from the wording of Art. 4. The purpose is different from a grade: the quiz makes it verifiable which content of the measure under Art. 4 has reached participants, instead of merely evidencing that it took place.

Sample questions:

• "Which of the following is a prohibited practice under Art. 5?" (multiple choice)
• "When do you have to label a chatbot?" (single choice)
• "Which data must NOT be entered into ChatGPT?" (multi-select)

Retest if not passed + after 12 months.

Practitioner recommendation on design: distribute the questions across the eight modules so that every module carries at least two questions — otherwise no gap can be traced back from the result. Map the three question types to their substantive suitability: single choice for allocations with exactly one correct provision, multiple choice for distinctions between risk classes, multi-select for data categories, because there several answers apply at the same time. The 12-month repetition is a practical rhythm, not a statutory deadline — section 7 places it in context.

6. Training record: content and retention

The training record documents who completed which module and passed the quiz, and when. The following structure is the practitioner recommendation from the first version; the wording of Art. 4 quoted in section 1 requires measures that support the development of AI literacy and prescribes no form of record for this — according to the Commission FAQ (non-binding), no certificate is required and an internal training record suffices.

Content of the record (practitioner recommendation from the first version):

• employee name + position
• training date
• module completions with time stamp
• quiz result
• certificate ID
• refresher date

Retention: for as long as the employee is with the company + 3 years after they leave.

Where the record connects legally beyond Art. 4 differs from provision to provision — the following table names, for each row, the provision and the trigger that is evidenced for precisely that provision; the last row summarises what remains without a source reference.

The retention rule 'for as long as they are with the company plus three years' is likewise a practical value. When handling the record, note that it contains personal data: it needs a legal basis Art. 6(1) GDPR, and requests for access or erasure must be answered without undue delay, at the latest within one month; the period can be extended by two months where the request is complex Art. 12(3) GDPR.

7. Introduction and repetition: rhythms in the business

The three usual triggers for a repetition — onboarding, annual refresher, ad hoc training — are practitioner recommendations intended to give substance to the measures-based yardstick of Art. 4 AI Act in the version applicable since 27 July 2026; the provision does not prescribe a repetition interval. They are expressly not statutory deadlines and must not be described as such in internal communication either.

This must be kept strictly separate from the training obligation under the NIS2 implementing law: it has a different subject matter (cybersecurity), a different group of addressees (management bodies of covered entities) and its own, statutorily ordered regularity. In Germany, the management bodies of particularly important entities and important entities must implement the risk management measures, supervise their implementation and take part in training on a regular basis § 38(1) to (3) of the German BSI Act (BSIG). In Austria, from the start of application of the Austrian Network and Information Systems Security Act 2026 (NISG 2026) on 1 October 2026: the management bodies of essential entities and important entities must ensure and supervise compliance with the risk management measures and take part in specifically designed cybersecurity training; the entities must offer their employees corresponding training on a regular basis § 31(1) and (2) NISG 2026. Anyone subject to both obligations should keep both records separately — that is a practitioner recommendation, not a legal consequence.

8. Germany and Austria

The AI Act is an EU regulation and applies directly and with identical wording in both countries — Art. 4 does not need to be transposed nationally, and there are no diverging national provisions. Only enforcement differs: each Member State establishes at least one notifying authority and at least one market surveillance authority, designates one market surveillance authority as single point of contact, notifies the Commission of this and makes the contact details publicly available Art. 70(1) and (2) AI Act; the penalties for breaches of Art. 4 are laid down by each Member State itself Art. 99(1) AI Act. In Germany, the market surveillance authority has been the Federal Network Agency (Bundesnetzagentur) since 29 July 2026 (§ 2(1) of the German AI Market Surveillance and Innovation Promotion Act (KI-MIG)), in so far as sectoral authorities — such as BaFin — are not competent; under § 6(1) KI-MIG, the single point of contact is likewise the Federal Network Agency (Bundesnetzagentur). As at the status of this article (17 August 2026), Austria has not designated a market surveillance authority under Art. 70 AI Act; the AI Service Desk at RTR is an information and advisory body without supervisory powers.

The practically relevant differences arise in the accompanying national law, which shapes the training content for the HR, IT and management personas:

For the management persona, the NIS2 classification is added, and only for covered entities. In Germany, another entity qualifies as a particularly important entity if it belongs to a type specified in Annex 1 and employs at least 250 people or has an annual turnover of more than EUR 50 million and, in addition, an annual balance sheet total of more than EUR 43 million; an entity qualifies as an important entity if it is of a type specified in Annex 1 or Annex 2 with at least 50 employees or with an annual turnover and an annual balance sheet total each exceeding EUR 10 million. Covered irrespective of size are, among others, operators of critical installations, qualified trust service providers, top-level domain name registries, DNS service providers and certain providers of telecommunications services § 28(1) and (2) BSIG. In Austria, from the start of application of the NISG 2026 on 1 October 2026, essential entities include — irrespective of size — qualified trust service providers, TLD name registries, DNS service providers and public administration at federal level, as well as large undertakings of the Annex 1 types; important entities are entities of the Annex 1 and Annex 2 types that operate a large or medium-sized undertaking, public administration at Land level, and, irrespective of size, telecommunications providers and trust service providers, in so far as they are not already essential entities § 24(1) and (2) NISG 2026. The size classes of the NISG 2026 (from 1 October 2026): a large undertaking from 250 employees or more than EUR 50 million annual turnover with more than EUR 43 million balance sheet total, a medium-sized undertaking from 50 employees or more than EUR 10 million turnover with more than EUR 10 million balance sheet total § 25(2) and (3) NISG 2026.

9. Penalties and liability risks

Breaches of Art. 4 have no separate EU range of fines: the catalogue in Art. 99(4) AI Act does not list Art. 4, and sanctioning follows Member State law Art. 99(1) AI Act. However, that only puts the question of penalties into perspective for Art. 4 itself — the content that the curriculum conveys is subject to penalties.

Every row of this table names a statutory maximum amount for precisely the infringement designated in the first column — these are not expected values. For SMEs and start-ups, the lower of the two values applies under Art. 99(6) AI Act; that mitigates the range, not the obligation. Added to this are the financial consequences of discrimination liability from section 8, which apply irrespective of the AI Act deadlines — for the HR persona, that is regularly the closer risk than the AI Act range of fines.

10. Provider comparison (DACH 2026)

The following comparison describes pricing model, content and legal coverage as at the first version of 2 May 2026 and was not re-surveyed in this update; specific prices and scopes of services are to be obtained from the provider before procurement.

Practitioner recommendation on selection: test a course against three questions — does it cover Module 5 with the eight Annex III areas, does it separate provider obligations from deployer obligations, and does it present Germany and Austria separately rather than as a DACH blend?