Why the AI literacy duty affects more businesses than many think
Since 2 February 2025, Article 4 of Regulation (EU) 2024/1689 (the AI Act) has required companies that provide or use AI systems to ensure a sufficient level of AI literacy among their staff. When the Council gave its final approval to the Digital Omnibus on 29 June 2026, many companies concluded that this duty had been scrapped, not least because the Commission’s original proposal from November 2025 would have moved the responsibility away from companies entirely. The final text does no such thing. The obligation was softened, not removed.
The timing also matters. From 2 August 2026, national market surveillance authorities begin supervising the AI Act, which turns a rule many companies have quietly ignored into something a regulator can actually ask about.
What does Article 4 require?
Article 4 covers both companies that develop AI systems and companies that merely use them, which in practice means almost every business. Anyone letting staff work with a customer service chatbot, a CV screening tool or a generative assistant is expected to make sure those staff understand what the technology can and cannot do, the risks it creates and how to read its output critically. The original wording asked companies to ensure literacy to the best of their ability, taking into account the staff’s technical knowledge, experience and training as well as the context in which the systems are used and the people affected by them.
What changes with the Digital Omnibus?
The final compromise text, endorsed by the European Parliament on 16 June 2026 and approved by the Council on 29 June 2026, rewrites Article 4 as an obligation of effort rather than result. Companies must take appropriate measures to support the development of AI literacy among staff and others operating AI systems on their behalf, but they no longer need to guarantee any measurable level of knowledge.
The Commission and the member states are tasked with supporting these efforts, and the European AI Board will issue recommendations with common objectives. The amending act has not yet been published in the Official Journal of the EU, so the new wording should be treated as provisional until publication, which is expected before 2 August 2026.
What counts as appropriate measures?
The duty is proportionate by design. What is appropriate depends on the systems in use, the risks they create and the existing knowledge of the people involved. There is no certification requirement and no mandated training format. The Commission’s published questions and answers on AI literacy point towards tailored measures: basic awareness for everyone who touches AI tools, and deeper training for the staff who select, configure or oversee systems. The Commission also maintains a living repository of AI literacy practices where companies can compare their approach with others.
A practical example: an accounting firm
A mid-sized Swedish accounting firm lets its consultants use a generative AI assistant for drafting reports and a specialised tool for detecting anomalies in client ledgers. An appropriate programme could consist of a short general session for all staff on what the tools may and may not be used for, targeted training for the consultants on hallucination risks and client confidentiality, and clear instructions for the team leads who review AI-assisted work. Documented attendance and updated guidelines when tools change complete the picture. The firm has not guaranteed any level of knowledge, but it can show appropriate measures, which is exactly what the revised Article 4 asks for.
Common mistakes companies make
The most common mistake right now is treating the softened wording as an abolition and cancelling planned training. Another is buying a single generic e-learning course for the entire workforce, which ignores the requirement that measures fit the context and the people. Many companies also train only their technical staff, although marketing, HR and finance teams often use AI tools more frequently. Finally, few document their efforts, and an undocumented effort is hard to demonstrate when a supervisory authority, a client or a court asks for it.
Recommended actions
Start by listing which AI systems are used in your business and by whom, including tools that individual teams have adopted on their own initiative. Group your staff by how they encounter AI and define what each group needs to understand. Run short, role-based training rather than one generic course, and repeat it when tools or rules change. Write down what you did, when and for whom, and keep the records with your other compliance documentation. Appoint someone responsible for keeping the programme alive, and follow the AI Board’s coming recommendations and the Commission’s repository of practices for inspiration.
Frequently asked questions
Can we be fined for insufficient AI literacy?
The AI Act’s penalty catalogue contains no specific fine for breaches of Article 4. Indirect risks remain: from 2 August 2026 national authorities supervise the regulation, and poor AI literacy can aggravate other breaches, undermine required human oversight and weigh against a company in liability disputes.
Does the duty apply to small businesses?
Yes. Article 4 contains no size threshold, but the duty is proportionate, so a small company with limited AI use can meet it with modest measures, such as clear usage instructions and a short training session, documented in writing and refreshed when tools change.
Was the AI literacy duty abolished by the Digital Omnibus?
No. The final text keeps the duty but reframes it: companies must take appropriate measures to support the development of AI literacy, an obligation of effort rather than result. The change takes legal effect once the amending act is published in the Official Journal.
Conclusion
The AI literacy duty is the AI Act’s quietest obligation, but it already applies to nearly every business and supervision begins on 2 August 2026. The Digital Omnibus lowers the bar from guaranteed knowledge to appropriate effort, which makes compliance realistic even for small organisations, provided the effort is genuine and documented. Companies that map their AI use, train by role and keep records turn a vague duty into a manageable routine, and often get better AI decisions in the bargain. At Lawgent, we help companies design proportionate AI literacy programmes, from mapping and role analysis to guidelines and documentation. Get in touch if you want to know what Article 4 means for your organisation.
