Guide · As of 19 October 2026
AI literacy obligation. What the EU AI Act requires today.
Since February 2025, companies that use AI have had to look after the AI literacy of their people. In July 2026 the rule was weakened. Here's what applies today and how to meet it without much effort.
By Andre Geimer, AI and data consultant
In shortOctober 2026
- The obligation still applies to every company that uses AI.
- What's required are measures such as training, no longer a specific level.
- No certificate needed, neither for trainers nor participants.
- In Germany there is no fine for it.
- Supervision and liability remain. Documentation helps.
What changed in July 2026.
The EU amended Article 4 with the so-called Digital Omnibus, Regulation (EU) 2026/1744. The core in two sentences:
Companies must, to their best extent, ensure a sufficient level of AI literacy.
Companies must take measures to support AI literacy. A specific level is explicitly not required.
How it came about
Article 4 applies. Companies are to ensure a sufficient level of AI literacy among their staff.
The European Commission proposes shifting the obligation from companies to the Commission and member states.
Parliament and Council agree on a middle way: the obligation stays with companies but is weakened.
The amendment is published as Regulation (EU) 2026/1744 in the Official Journal and applies.
National authorities take over supervision. In Germany that's mainly the Federal Network Agency (Bundesnetzagentur).
Is there a fine?
No. But it isn't without consequences.
The AI Act does not list Article 4 in its penalty provisions. Germany's implementing act, in force since 29 July 2026, doesn't set one either.
The European Commission points out that supervisory authorities can sanction breaches of Article 4 with other enforcement measures too. In Germany that's mainly the Bundesnetzagentur.
According to the Bundesnetzagentur, a lack of AI literacy can count as a breach of the duty of care, especially if damage results. If someone leaks confidential data through an AI tool, documented training is a good argument.
What training should cover.
The easiest way to meet the obligation is training. The Bundesnetzagentur suggests three levels for it. They're not binding, but a good benchmark.
- Level 1Basics
Terms, how AI works, which technologies exist, opportunities and risks.
- Level 2Your use
Your role as a company, the AI tools you use, their risks and the legal classification.
- Level 3By role
Deeper training for specific tasks, such as technology, law or ethics.
Not required
- A certificate for participants
- A certified trainer or an external provider. Internal training counts too.
- A standardised training
- An AI officer
Recommended
- Tailor content to roles and risks
- More than pointing to the tools' instructions
- Document which measure, what scope, who took part
My "AI basics" workshop covers the first two levels, on your own tools and tasks, with a certificate of attendance for your records.
See the workshopFive steps to meet it.
No officer, no certificate, done in a few weeks.
- Take stock
Which AI tools do your people use, officially and unofficially?
- Set rules
What may go into which tool, and what may not? Start with customer data, personnel data and trade secrets.
- Train by role
Someone who uses AI occasionally needs different content from someone who prepares decisions with it. Ideally on their own tasks.
- Document
Who was trained when and on what? A simple list is enough.
- Refresh
Tools and models change quickly. A refresher once a year is a good rhythm.
Is there funding?
In Germany, the training subsidy of the Federal Employment Agency usually doesn't apply to short courses of one or two days, because it requires more than 120 hours.
In Saarland, the state programme "Kompetenz durch Weiterbildung" covers 40 percent of seminar costs for small and medium-sized companies based in Saarland, up to €2,000 per employee. It funds open seminars by external providers, not training run exclusively for one company. Whether it currently has funds is best checked directly with the programme office.
How I can help.
In my AI workshops your team learns to work with AI on its own tasks and agrees clear rules for using it. That is a measure in the sense of Article 4, and you get a certificate of attendance for your records.
Frequently asked questions.
Is AI training for employees mandatory?
Yes, for companies that use or provide AI systems. Since July 2026 they must take measures that support the AI literacy of their staff. A specific level is not required.
Does the training have to be certified or delivered by a certified trainer?
No. According to the Bundesnetzagentur, the AI Act does not require certification of training, and participants don't need a certificate either. What matters is that content fits roles and risks and is documented.
Does internal training count?
Yes. The Bundesnetzagentur states explicitly that measures can be carried out internally or externally. An external workshop is mostly worth it if nobody in-house has the time or experience to prepare the content.
Is there a fine if we don't train our staff?
Not in Germany. Neither the AI Act nor the German implementing act sets a fine for Article 4. Supervisory authorities can take other measures, though, and a liability risk remains if untrained staff cause damage with AI.
Since when does the obligation apply?
Since 2 February 2025. In July 2026 it was weakened by Regulation (EU) 2026/1744.
Sources
- European Commission, AI literacy questions and answers (updated 27 July 2026)
- European Parliament, Legislative Train, Digital Omnibus on AI
- Council of the EU, text of the amending regulation (PE-CONS 30/26)
- Bundesnetzagentur, AI literacy (German)
- Bundesnetzagentur, press release on the implementing act (29 July 2026, German)
- KI-Marktüberwachungsgesetz, § 15 penalty provisions (German)
- § 82 SGB III, funding for employed workers (German)
- Förderdatenbank, Kompetenz durch Weiterbildung, Saarland (German)
This guide reflects the situation as of 19 October 2026 and is not legal advice.