Compliance & funding
AI Literacy Obligation & AI Policy Template 2026
AI literacy training under Article 4 AI Act after the Omnibus: who is covered, enforcement, a role-based training plan, record template and an AI policy template.

Key takeaways
In short:
- Yes, there is an AI literacy obligation. Under Article 4 of the AI Act, every provider and deployer of AI systems has had to take AI literacy measures for its staff since 2 February 2025. Using ChatGPT for marketing copy is enough to be in scope.
- Since 27 July 2026, the softer Omnibus wording applies: organisations must "support the development of" AI literacy and do not have to guarantee any specific level for any individual.
- No certificate, no AI officer, no dedicated EU fine, but measures that fit roles and risks, and records to show them. National market surveillance authorities have supervised the rules since 2 August 2026.
- A workable proof of compliance has three parts: a role-based training plan, an AI acceptable use policy and training records. You will find templates for all three below.
Many articles on AI literacy still quote the original legal text or warn of multi-million-euro fines that do not exist for Article 4. Others explain the law but stop where implementation begins. This guide reflects the position in November 2026: what Article 4 requires after the AI Omnibus, who it covers, how it is enforced across the EU, and how to get to solid evidence with a training plan, a policy template and a 90-day roadmap.
This article is not legal advice. It does not replace a review of your specific use of AI by a qualified lawyer in your jurisdiction.
What does Article 4 require after the Omnibus?
AI literacy is defined in Article 3(56) of the AI Act as the skills, knowledge and understanding that allow providers, deployers and affected persons to make an informed deployment of AI systems and to be aware of the opportunities, risks and possible harm of AI. Article 4 turns this into a duty for organisations. Its wording changed in 2026:
| Original wording (2 Feb 2025 to 26 Jul 2026) | Wording since 27 Jul 2026 (AI Omnibus) | |
|---|---|---|
| Core duty | take measures "to ensure, to their best extent, a sufficient level of AI literacy" of staff | take measures "to support the development of AI literacy" of staff |
| Level | "sufficient level" as the target | explicitly no specific level has to be guaranteed for any individual |
| Yardstick | technical knowledge, experience, education and training, context of use, persons affected | unchanged |
| Who must act | providers and deployers | unchanged |
| Role of the EU and Member States | not addressed | Commission and Member States support organisations, in particular SMEs; Commission publishes practical examples; AI Board adopts recommendations |
Sources: original text in Regulation (EU) 2024/1689; new text in the consolidated Article 4 as amended by Regulation (EU) 2026/1744. The European Commission's AI literacy Q&A sums it up: AI literacy remains an obligation, but no specific or "sufficient" level is mandated any more.
In practice, you owe appropriate measures, not a guaranteed learning outcome for every person. If you have trained people demonstrably, set rules and answered their questions, you are not responsible for an employee forgetting the content. If you have done nothing, you are still in breach of applicable law.
The stricter exception: high-risk AI
Deployers of high-risk AI (for example CV screening or consumer credit scoring) must also meet Article 26(2): human oversight must be assigned to people with the necessary competence, training and authority. This duty was not softened, as the Commission stresses. For Annex III systems it applies from 2 December 2027; see our EU AI Act timeline 2026–2028 for details.
Who is covered by the AI literacy obligation?
Article 4 applies to providers (who develop an AI system or put it into service under their own name) and deployers (who use an AI system under their authority in a professional capacity). Sector, company size and risk class do not matter. Germany's regulator, the Bundesnetzagentur, explicitly includes general-purpose systems such as chatbots.
In concrete terms, the duty covers:
- Your own staff who operate AI systems or use their output, including part-timers, working students and apprentices.
- Other persons acting on your behalf. The Commission gives contractors, service providers and even clients as examples: people who are not employees but act within your organisational remit.
- Everyday use of off-the-shelf tools. Asked whether a company whose staff use ChatGPT for advertising copy or translation must comply with Article 4, the Commission answers: yes, they should be informed about specific risks such as hallucinations.
Purely personal use is out of scope. But when employees use private AI accounts for work tasks, that is professional use within your remit. This is shadow AI, and it is the reason an AI policy belongs to any Article 4 programme.
What Article 4 does not require
According to the Commission's Q&A and the Bundesnetzagentur, the following are not required:
- a certificate or external certification of the training,
- a standardised training format or minimum number of hours,
- measuring or testing employees' knowledge,
- an AI officer or AI governance board,
- sector-specific requirements.
Anyone selling an "AI licence" as a legal requirement is overstating the case. A certificate course can still make sense for power users, as a qualification rather than a legal duty.
What happens if you do nothing?
Article 4 is enforced not by the EU AI Office but by national market surveillance authorities, from 2 August 2026 onwards (Commission). Each Member State designates its own authorities; in Germany it is primarily the Bundesnetzagentur under section 2 of the KI-MIG (Federal Law Gazette 2026 I No. 223).
The sanctions picture as of November 2026:
| Level | Rule for Article 4 | What it means |
|---|---|---|
| AI Act, Article 99(3)–(5) | Article 4 is not listed in any of the three fine tiers (EUR 35m / 15m / 7.5m) | no EU-wide fine range for missing training |
| AI Act, Article 99(1) | Member States lay down penalties "and other enforcement measures", which may include warnings and non-monetary measures, for any infringement | national sanctions are possible in principle; check your country's implementing law |
| Germany, section 15 KI-MIG | fines up to EUR 50,000 for breaches of Articles 21, 27, 45 and 86, not Article 4 | no German fine for Article 4 |
| Liability and duty of care | according to the Bundesnetzagentur, a lack of AI literacy can be regarded as a breach of the duty of care, especially when harm occurs | relevant for damages claims, organisational fault and director liability |
| Other laws | GDPR, trade secrets, professional secrecy, copyright | this is where real fines and losses arise |
The Commission itself says sanctions must be proportionate and become more likely if there is proof of an incident caused by a lack of training. That is the realistic scenario: an employee uploads a customer list to a personal AI account, a salesperson sends a quote containing hallucinated terms, a recruiter lets an unvetted tool pre-sort applications. The first question will be what rules and training were in place. Regulated professions such as lawyers, accountants and doctors also face professional secrecy rules in every Member State, regardless of the AI Act.
What does a role-based training plan look like?
Neither the Commission nor national regulators prescribe a curriculum. The Commission lists minimum considerations: a general understanding of AI, your role as provider or deployer, the risks of the systems you use, and targeted measures built on that analysis. The Bundesnetzagentur recommends a tiered approach: level 1 basics for everyone, level 2 advanced knowledge of the systems in use, level 3 role-specific training.
This leads to the plan below. Durations are our recommendation for a mid-sized company with typical tools (a chat assistant, AI in office software, AI features in CRM or helpdesk), not a legal requirement.
| Target group | Content (on top of the baseline) | Format | Scope (recommended) |
|---|---|---|---|
| All staff (baseline) | How language models work, hallucinations, data classes, approved tools, labelling, incident reporting, the AI policy | e-learning or live session with exercises on real tasks | 2–4 hours, annual refresher |
| Management | Roles under the AI Act, risk classes, liability, tool and budget decisions, leading by example, dealing with shadow AI | short briefing or workshop | half a day |
| Power users, AI champions | Prompting techniques, checking facts and sources, automation, tool limits, acting as internal contact | one-day seminar or multi-day course | 1–3 days |
| IT, developers, data protection | Provider vs deployer role (Article 25), API use, data flows, processor agreements, prompt injection, logging, assessing new tools | specialist workshop | 1 day |
| HR, recruiting | High-risk classification under Annex III point 4, banned emotion recognition at work, human oversight, discrimination risks, employee representatives | workshop with case studies | half a day; mandatory before any HR AI tool goes live |
| Marketing, sales | Copyright and licences for AI output, Article 50 labelling (deepfakes), fact and brand checks, customer data in CRM | hands-on workshop | half a day |
| Customer service | AI disclosure for chat and voice bots, hand-over to humans, complaints about AI answers, no commitments without checking | live training with role play | 2–4 hours |
| External parties (freelancers, agencies) | Your AI policy, data classes, labelling | policy as a contract annex, short briefing | 30–60 minutes |
Two practical points: the Commission considers merely forwarding instructions for use often ineffective. And staff with an AI degree are "normally" AI literate, but still need to know your specific systems, risks and legal aspects.
Free support for SMEs
Article 4(2) obliges the Commission and Member States to support organisations, in particular SMEs. Available today:
- the Commission's living repository of AI literacy practices, with more than 40 examples according to the Q&A; copying them does not automatically create a presumption of compliance,
- the European Digital Innovation Hubs (EDIHs), according to the Commission more than 200 one-stop shops across Europe, 9 out of 10 of which already offer AI services such as training and workshops,
- the EU's AI Skills Academy, operating since 1 May 2026 (Commission),
- national programmes, for example Germany's Mittelstand-Digital centres with around 100 AI trainers according to the Bundesnetzagentur.
These offers mainly cover the basics. Training on your tools, data and processes is still up to you. For German funding programmes that subsidise training, see our guide to digitalisation and AI funding in Germany.
How do you document compliance?
The Commission requires no certificate; an internal record of training and other initiatives is enough. The Bundesnetzagentur recommends recording the type of measure, its content and duration and the participants, evaluating measures regularly and refreshing them when needed. A spreadsheet or folder will do. It should contain these fields:
| Field | Example |
|---|---|
| Measure and format | Live training "AI basics and policy", remote |
| Date, duration | 12 Nov 2026, 3 hours |
| Trainer or provider | internal: head of IT / external: training provider |
| Target group and participants | Sales, 9 people (list with name, role, signature or click confirmation) |
| AI systems covered | Chat assistant (business account), AI features in CRM |
| Content in keywords | Hallucinations, data classes, labelling, incident reporting |
| Link to risk classification | minimal risk / transparency obligation; no high-risk |
| AI policy version | v1.1 of 1 Nov 2026, acknowledged |
| Effectiveness check (optional) | Short quiz, feedback, spot check after 4 weeks |
| Next refresher, owner | November 2027 or when a new tool is introduced; sales team lead |
Add two things that are asked first in an authority request or after an incident: an up-to-date AI inventory (which systems, for what, who uses them) and an onboarding step that trains new joiners automatically.
Template: what belongs in an AI acceptable use policy?
An AI policy (also AI acceptable use policy or AI usage guideline) is the binding internal rule on which AI tools may be used by whom, for what, with which data, and who is accountable for the results. It is not a legal requirement in itself. But it is the simplest way to bring Article 4, the GDPR and trade-secret protection together in one document. Good policies are approval rules, not lists of bans: if you only prohibit, you get shadow AI.
Adapt the structure and sample clauses below. Placeholders are in square brackets. Organisations with employee representatives, high-risk use or professional secrecy obligations should have it reviewed by a lawyer.
1. Purpose and scope
This policy governs the professional use of AI systems at [Company]. It applies to all employees and to freelancers, service providers and agencies working on our behalf, regardless of device, account or location.
2. Approved tools
Only the AI tools listed in Annex 1 may be used for work, and only via company accounts. Personal accounts must not be used for work tasks. Request new tools informally from [role]; data protection, contract and security are reviewed within [10] working days.
Annex 1 lists for each tool: name, purpose, permitted data class, account type, contact person, status of the data processing agreement under Article 28 GDPR and whether inputs are used to train the vendor's models.
3. Data classes: what may go where?
| Class | Examples | Rule |
|---|---|---|
| Public | published website copy, general professional questions | any approved tool |
| Internal | concepts, process descriptions, draft quotes without personal data | only tools with company account and contract |
| Confidential | personal data of customers and staff, financial figures, contracts | only systems explicitly approved for this, e.g. a self-hosted AI |
| Never enter | passwords and API keys, health and applicant data, professional secrets, NDA data, security-relevant source code | no external AI tool, not even as an excerpt or screenshot |
If in doubt, apply the higher class. Anonymise data before input wherever possible.
4. Reviewing output
AI output is a draft. Whoever uses it is fully responsible for it. Figures, quotes, sources and legal statements are checked before any external use. Decisions about people (applications, appraisals, dismissals, creditworthiness) are always made by a human; AI may only assist here after specific approval.
5. Labelling
We visibly label image, audio or video content that realistically depicts real people, places or events (deepfakes) as AI-generated. Chat and voice bots tell users at the start that they are interacting with AI. Editorially reviewed text is labelled according to [internal rule].
The background is Article 50 of the AI Act, applicable since 2 August 2026. For text published to inform the public on matters of public interest, the disclosure duty does not apply if it has undergone human review and someone holds editorial responsibility. Our upcoming guide to Article 50 labelling covers wording and examples.
6. Copyright and trade secrets
Do not enter third-party protected content (text, images, code) into AI tools without the rights to do so. Before publication, check whether AI output resembles existing works or trademarks. Follow each tool's licence terms on commercial use.
7. Prohibited uses
In particular, the following are not permitted: emotion recognition of employees, covert AI-based monitoring of performance or behaviour, generating intimate or defamatory content about real people, and any use that breaches Article 5 of the AI Act.
8. Shadow AI and browser extensions
AI features in browser extensions, meeting note-takers and plug-ins count as separate tools and require approval. Anyone who notices unapproved use reports it to [role] without negative consequences.
9. Incident reporting
If data classed as "Confidential" or "Never enter" has been entered into an unapproved tool by mistake, report it to [data protection/IT] immediately. Speed matters more than blame.
The reason: a personal data breach may have to be notified to the supervisory authority within 72 hours under Article 33 GDPR. A culture where mistakes are hidden costs more than the mistake.
10. Training
All staff complete the baseline training before first use and refresh it annually. Certain roles take additional modules according to the training plan. Attendance is recorded.
11. Responsibilities
[Role] owns this policy. Tools are approved by [IT] together with [data protection]. Departmental AI champions are the first point of contact. The policy is reviewed at least [every six months] and whenever new tools or legal changes arise.
12. Employee representatives
Many EU countries give works councils or employee representatives information, consultation or co-determination rights when technology that can monitor staff is introduced, so check your national rules early. In Germany, the Hamburg Labour Court (decision of 16 January 2024, 24 BVGa 1/24) found no co-determination right where staff were merely permitted to use ChatGPT via their own browser accounts. If a company rolls out AI tools itself, with company accounts and logs, section 87(1) no. 6 of the Works Constitution Act (BetrVG) may apply because such tools can be suitable for monitoring behaviour or performance. The employer must also inform the works council in good time about the use of AI under section 90 BetrVG. A framework works agreement on AI that mirrors the policy and handles new tools via an annex works well in practice.
How do you roll this out in 90 days?
| Period | Steps | Result |
|---|---|---|
| Days 1–30 | Name an owner; build an AI inventory including shadow AI (a short anonymous survey helps); record role and risk class per system; draft the AI policy; inform employee representatives | inventory, policy v0.9, schedule |
| Days 31–60 | Approve or replace tools (company accounts instead of personal ones); adopt the policy; baseline training for everyone; management briefing | policy v1.0, baseline attendance list |
| Days 61–90 | Role modules (HR, marketing, service, IT); train power users; add an onboarding step; check labelling for chatbots and deepfakes; first evaluation | complete records, refresher plan |
Checklist: is your Article 4 evidence complete?
- An up-to-date AI inventory lists purpose, user groups and risk classification.
- The AI policy is adopted, communicated and acknowledged by everyone.
- Everyone who uses AI at work has completed the baseline training.
- Higher-risk roles (HR, marketing, service, IT) have taken additional modules.
- External parties working with AI on your behalf are bound to the policy by contract.
- New joiners are trained during onboarding.
- Attendance, content and policy version are recorded, and the next refresher is scheduled.
- There is a known reporting channel for AI incidents.
Conclusion: softer, but not gone
The AI Omnibus softened the AI literacy obligation; it did not abolish it. For most organisations that is good news: instead of an unclear "sufficient level", what counts now is whether you have taken traceable measures that fit roles and risks. With an inventory, a two- to six-page policy, baseline training for everyone and extra modules for HR, marketing, service and IT, this is achievable within a quarter, and your team ends up not only more compliant but also more productive with AI.
If you would rather not set this up alone: our AI training ranges from an online course (EUR 500 per person) through a one-day seminar (EUR 600 per person) and a certificate course for power users (EUR 2,300 per person) to in-house training with your own tools and examples (EUR 2,000 for a half day, EUR 3,500 for a full day per team); a strategy and use-case workshop costs EUR 4,500 per day. All prices include German VAT. For inventory, risk classification and aligning the policy with the GDPR, our privacy, GDPR and EU AI Act consulting is available at a day rate of EUR 1,200 (incl. VAT); it does not replace legal advice. Get in touch for a no-obligation conversation.
Sources
All sources accessed 6 November 2026.
- AI Act, consolidated version (as of 27 July 2026), Article 4 AI literacy: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-4
- AI Act, consolidated version, Articles 3(56), 26, 50 and 99 (same address pattern): https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-99
- Regulation (EU) 2024/1689 (AI Act), original version, Official Journal L of 12 July 2024: https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=OJ:L_202401689
- Regulation (EU) 2026/1744 (AI Omnibus), Official Journal L of 24 July 2026: https://eur-lex.europa.eu/legal-content/DE/TXT/?uri=OJ:L_202601744
- European Commission, AI Literacy – Questions & Answers: https://digital-strategy.ec.europa.eu/en/faqs/ai-literacy-questions-answers
- European Commission, Living repository of AI literacy practices: https://digital-strategy.ec.europa.eu/en/policies/repository-ai-literacy-practices
- Bundesnetzagentur (German market surveillance authority), AI Service Desk, "KI-Kompetenz" page: https://www.bundesnetzagentur.de/DE/Fachthemen/Digitales/KI/7_Kompetenz/start.html
- German Federal Law Gazette, BGBl. 2026 I No. 223 (AI Act implementing law incl. KI-MIG): https://www.recht.bund.de/bgbl/1/2026/223/VO.html
- Regulation (EU) 2016/679 (GDPR), Articles 28 and 33: https://eur-lex.europa.eu/eli/reg/2016/679/oj/eng
- German Works Constitution Act (BetrVG), sections 87 and 90: https://www.gesetze-im-internet.de/betrvg/__87.html
- LTO, report on Hamburg Labour Court, decision of 16 January 2024, 24 BVGa 1/24: https://www.lto.de/recht/hintergruende/h/arbg-hamburg-24bvga-1-24-chatgpt-einsatz-arbeit-beteiligung-betriebsrat
Frequently asked questions
Is AI literacy training mandatory under the EU AI Act?
Article 4 has required providers and deployers of AI systems to take AI literacy measures for their staff since 2 February 2025. The law does not prescribe a format. The European Commission notes, however, that simply forwarding instructions for use is often ineffective, so training is the usual way to comply.
What did the Digital Omnibus change in Article 4?
Since 27 July 2026, providers and deployers no longer have to ensure, to their best extent, a sufficient level of AI literacy. They must take measures to support its development, and the text states that no specific level has to be guaranteed for any individual. The obligation itself remains.
What is the fine for not providing AI literacy training?
Article 99 of the AI Act sets no specific fine for Article 4, and Germany's implementing act (KI-MIG) adds none either. Member States may still provide other enforcement measures, and the German regulator says a lack of AI literacy can be treated as a breach of the duty of care, especially when harm occurs. Real fines usually come from other laws such as the GDPR.
Do we need a certificate or an AI officer?
No. The Commission's Q&A says no certificate is needed and an internal record of training and other initiatives is enough. Article 4 does not require an AI officer or a governance board either, although naming one responsible person is good practice.
Does Article 4 apply if staff only use ChatGPT?
Yes. The Commission answers this exact question: a company whose employees use ChatGPT for advertising copy or translation must comply, and staff should be informed about specific risks such as hallucinations. Only purely personal, non-professional use is outside the scope.


