TLDR: The European Union softened its rule on staff training in artificial intelligence this summer, yet employers still have to support it, and Swiss firms with European teams should prove it role by role.
Regulation (EU) 2026/1744 turned the Article 4 duty to ensure literacy into a duty to support it
The European Union (EU) Artificial Intelligence Act, Regulation (EU) 2024/1689, has carried an artificial intelligence (AI) literacy obligation since its first provisions applied on 2 February 2025, under Article 113(a). In July 2026 the EU rewrote it. Regulation (EU) 2026/1744 of 8 July 2026, known as the Digital Omnibus on AI, was published in the Official Journal on 24 July 2026, as Cuatrecasas reported the following day. Point 5 of its Article 1 replaces Article 4 in full.
The new text reads: “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”, taking into account their technical knowledge, experience, education and training, the context of use and the people on whom the systems are used. It then adds a sentence that the original lacked: “This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual.”
Recital 8 of the amending regulation explains the change. It records that the original Article 4 obliged providers and deployers “to ensure AI literacy of their staff”, and that this standard proved too demanding for many organisations, smaller enterprises in particular. The same recital keeps the policy aim in plain view: “AI literacy should be a strategic priority, regardless of regulatory obligations and potential sanctions”. Paragraphs 2 and 3 of the new Article 4 ask the European Commission and member states to support employers, particularly small and medium-sized enterprises, and the European Artificial Intelligence Board to adopt recommendations based on European competence frameworks.
The education platform KI-Campus summarised the outcome in one line: Article 4 “will not be removed, but instead pragmatically recalibrated”, and providers and deployers “do not have to guarantee” that every individual reaches a certain level of competence. The duty to act remains. What changed is the standard an employer is measured against: from a result to an effort.
What does Article 4 now ask of an employer that uses artificial intelligence?
An employer that uses an AI system at work is a deployer in the language of the Act. Article 3(4) defines a deployer as “a natural or legal person, public authority, agency or other body using an AI system under its authority”, with an exception for personal, non-professional use. A pharmaceutical company whose medical information team drafts replies with a generative assistant, a medtech firm whose sales team writes call notes with one, and a bank whose human resources (HR) team screens internal applications with one are all deployers.
Article 3(56) defines the target skill. AI literacy means “skills, knowledge and understanding that allow providers, deployers and affected persons”, taking into account their rights and obligations, “to make an informed deployment of AI systems, as well as to gain awareness about the opportunities and risks of AI and possible harm it can cause”. Read together with the new Article 4, the obligation has four working parts:
- Who: staff and other persons who operate or use AI systems on the employer’s behalf, which can include contractors and service providers.
- What: measures that support the development of AI literacy, such as training, guidance and supervised practice.
- How much: proportionate to each person’s knowledge, experience and the context in which the system is used.
- For whom: calibrated to the people on whom the system is used, such as patients, customers or job applicants.
The law sets the duty; how an employer meets it in practice is a matter of judgement. Employers should take advice from counsel in each EU country where they operate on how the national authorities will read their own training and use policies.
The European Commission accepts an internal training record as evidence
The Commission’s own guidance, published on 7 May 2025 for the original text, remains the most practical reading of what an employer must show. Its questions and answers on AI literacy state: “There is no need for a certificate.” The same page adds that “Organisations can keep an internal record of trainings and/or other guiding initiatives.” It also treats a common shortcut with scepticism: relying only on the instructions for use, or asking staff to read them, “might be ineffective and insufficient”.
The guidance also answers the question most employers ask first, about staff who use a general-purpose assistant. For employees who use ChatGPT to write advertising text or translate documents, the Commission’s answer is: “Yes, they should be informed about the specific risks, for example hallucination.” The amended Article 4 lowers the bar from ensuring to supporting, and that answer still describes the minimum an employer can reasonably document: people who use a tool know what it gets wrong.
Swiss employers with teams in the European Union fall inside the duty
The Act reaches beyond EU borders. Article 2(1)(b) covers “deployers of AI systems that have their place of establishment or are located within the Union”, and Article 2(1)(c) extends the Act to providers and deployers in third countries “where the output produced by the AI system is used in the Union”. A Swiss group’s subsidiaries in France, Germany or Italy are deployers in their own right. A Basel or Vaud team whose AI-assisted output, such as a marketing email or a medical information reply, is used in the EU may also fall within scope.
Switzerland is building its own framework on a slower clock. On 12 February 2025 the Federal Council decided to bring the Council of Europe’s AI convention into Swiss law, with amendments that “should be as sector-specific as possible” and a consultation draft planned by the end of 2026. Swiss business opinion leans towards the EU model: in EY’s 2026 survey of 604 Swiss companies, “40% favor alignment with the EU AI Act”, against 28% who support an independent Swiss approach. A Swiss employer that documents AI literacy to the EU standard today is preparing for the most likely direction of its home rules as well.
| Obligation | Applies from | Change in 2026 |
|---|---|---|
| Article 4 artificial intelligence (AI) literacy, all providers and deployers | 2 Feb 2025 | “Ensure” became “take measures to support” |
| Annex III high-risk systems, including recruitment and performance evaluation | 2 Dec 2027 | Moved from 2 Aug 2026 |
| Annex I high-risk AI inside regulated products | 2 Aug 2028 | Moved from 2 Aug 2027 |
| Swiss consultation draft on AI rules | Planned by end 2026 | Federal Council decision of 12 Feb 2025 |
The Omnibus moved the high-risk dates and left the literacy duty in force. Cuatrecasas notes that stand-alone Annex III systems, which include tools for recruitment, personnel management and performance evaluation, now face their obligations from 2 December 2027, sixteen months later than planned. Hiring software therefore has a later deadline, while the people who use every other AI tool fall under Article 4 already. Edward Galle covered the hiring side in its playbook on EU artificial intelligence rules for hiring software; this article concerns the staff who already use AI at work.
Swiss staff use artificial intelligence daily, and their training lags behind
Adoption in Switzerland is broad. EY’s survey found that “89% of respondents already use AI solutions in their daily work”. It also found that 29% report their company permits external AI tools through private accounts such as ChatGPT, and that “for 8%, this is currently the only way to access AI in their everyday work”. Asked which policy measure matters most, 58% named the expansion of education and training in AI.
Employees describe the same gap from the other side. PwC’s Hopes and Fears survey of 996 Swiss workers, fielded in May and June 2026, found that 28% use generative AI daily, against 22% globally, while only 19% strongly agree they understand the skills they will need in future, against 25% globally, and 23% perceive access to learning and development resources, against 26% globally. Swiss staff use the tools more than the global average and feel less equipped to use them well.
Private accounts matter for Article 4. In Edward Galle’s reading, a company that tolerates staff using a personal assistant for work tasks is still a deployer of that use if the system runs under its authority, and its literacy measures must reach those users. Edward Galle’s analysis of Swiss small and medium-sized enterprises that use AI while lacking rules on what staff may enter showed how often that use runs ahead of policy.
Field evidence shows that artificial intelligence improves some tasks and degrades others for the same user
The strongest case for role-based literacy comes from a field experiment. Fabrizio Dell’Acqua and co-authors studied 758 consultants of Boston Consulting Group (BCG), now published in Organization Science. On tasks inside the frontier of what AI does well, consultants with AI completed 12.2% more tasks and worked 25.1% faster. On a task outside that frontier, consultants using AI were 19 percentage points less likely to reach a correct solution than those working alone.
The study’s “jagged frontier” describes exactly what Article 4 asks employers to teach. A tool that improves one task and degrades the next looks identical in both cases to an untrained user. Literacy, in the Act’s own words, is the awareness of “opportunities and risks of AI and possible harm it can cause”, and in practice that means knowing which tasks sit on which side of the frontier for a given role.
The skills pressure is broad. The World Economic Forum’s Future of Jobs Report 2025 found that workers can expect “two-fifths (39%) of their existing skill sets” to be transformed or become outdated between 2025 and 2030, and 63% of employers named skill gaps as the main barrier to business transformation. Edward Galle’s earlier analysis of human resources as the fastest-growing field for AI skills in Swiss job ads tracked the same shift in what employers now ask of new hires. Article 4 asks the same of the people already on the payroll.
Literacy for commercial teams, medical information and human resources looks different
Edward Galle’s view, from searches for digital, commercial and HR roles in Swiss pharma, medtech and finance, is that a single generic course rarely satisfies the “context of use” test in the new Article 4. Each function meets AI in different systems, with different people on the receiving end.
Commercial and brand teams use generative tools to draft campaign copy, segment customer lists and summarise account notes. In pharma and medtech the output reaches healthcare professionals, so literacy covers promotional review rules, the risk of invented claims and the handling of customer data in prompts. The people affected are prescribers and patients, and the training should say so.
Medical information and pharmacovigilance teams use AI to triage enquiries, search literature and draft responses. Here literacy means knowing when an output needs a qualified person’s check before it leaves the building, and how to recognise a possible adverse event inside an AI-drafted summary. The context of use is a regulated process with an audit trail, so the training record belongs in that same trail.
Human resources teams sit on both sides of the Act. They use AI in their own work, from drafting job descriptions to answering employee questions, and they own the training programme for everyone else. Recruitment and performance tools fall under the Annex III high-risk rules from December 2027; until then, the HR team’s own literacy is what lets it challenge a vendor’s claims and explain a system to the employees it affects.
Three documents show a supervisor that staff are literate in artificial intelligence
Edward Galle recommends that a regulated employer be able to put three documents in front of an auditor, a works council or a client’s compliance team.
A training record by role. The record lists who uses which AI systems, what training or guidance each person received, and when. The Commission’s guidance accepts an internal record, so the format can be simple: a learning management system report or a spreadsheet maintained by HR. What makes it credible is the link between the role and the content, such as a medical information specialist trained on hallucination in literature summaries.
A role-based use policy. The policy states which tools each function may use, for which tasks, and with which data. It answers the private-account question explicitly, either by providing an approved tool or by setting out what staff may enter into a personal one. It is short enough for staff to read and specific enough for an auditor to test.
A named owner. One person, in a reporting line that HR administers, answers for literacy across the organisation and for each AI system in use. Edward Galle’s article on why every agent in a company needs a named human principal explains why that accountability belongs with HR. The owner reviews the record, updates the policy when a new tool arrives and reports to the executive committee.
The Commission accepts an internal record in place of a certificate, so all three documents can be built with existing HR systems. Learning programmes help most when they follow the same role logic. Kainjoo, the company behind Edward Galle, has partnered with Harvard Business Review to build training journeys for Edward Galle candidates and client teams, an example of literacy built around a career stage and a function.
Hiring managers can recruit for literacy as well as train for it
Article 4 also changes what a good hire looks like. A candidate who can explain where an AI tool fails in their own function, and how they checked its output in a previous role, reduces the training an employer must document. Edward Galle’s analysis of AI skills in Swiss marketing job ads showed employers already asking for these skills. Interview questions can test them directly: which task would you keep away from an assistant, and why? What did you do the last time a tool gave you a confident wrong answer?
Edward Galle’s view is that these questions belong in every brief for a role exposed to AI, and its specialist search and human resources consulting work for Swiss employers covers both sides of that hire. Candidates who can show that judgement can browse open positions or send a spontaneous application.
Swiss employers that document role-based literacy now will meet the EU duty and the 2027 high-risk rules from the same base
The Omnibus gave employers a lighter standard and a longer runway for high-risk systems. It also left one obligation already in force since February 2025 and confirmed it in new wording. A Swiss group with EU teams that keeps a training record by role, a role-based use policy and a named owner meets Article 4 as amended, answers the Commission’s guidance and holds the foundation the Annex III rules will require for recruitment and performance tools in December 2027.
This article has been reviewed by Haider Alleg, Managing Partner at Kainjoo SA.
References
- European Parliament and Council. Regulation (EU) 2026/1744 of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 (Digital Omnibus on AI). Official Journal of the European Union, 24 July 2026. https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202601744
- Cuatrecasas. The Digital Omnibus on AI: How does it impact employment relations? 25 July 2026. https://www.cuatrecasas.com/en/spain/labor-and-employment/art/digital-omnibus-on-ai-how-does-it-impact-employment-relations
- KI-Campus. EU AI Act: AI Omnibus recalibrates Article 4, AI literacy remains mandatory. 28 July 2026. https://ki-campus.org/en/blog/eu-ai-act-ai-omnibus-recalibrates-article-4-ai-literacy-remains-mandatory
- Future of Life Institute. EU Artificial Intelligence Act, Article 4: AI literacy (consolidated text). Accessed 3 November 2026. https://artificialintelligenceact.eu/article/4/
- Future of Life Institute. EU Artificial Intelligence Act, Article 3: Definitions. Accessed 3 November 2026. https://artificialintelligenceact.eu/article/3/
- Future of Life Institute. EU Artificial Intelligence Act, Article 2: Scope. Accessed 3 November 2026. https://artificialintelligenceact.eu/article/2/
- European Commission. AI Literacy: Questions and Answers. 7 May 2025. https://digital-strategy.ec.europa.eu/en/faqs/ai-literacy-questions-answers
- European Commission. AI Literacy: Questions and Answers (PDF copy hosted by Sidley Austin). 2025. https://www.sidley.com/en/-/media/resource-pages/ai-monitor/guidance/eu-commission-ai-literacy–questions-and-answers.pdf?la=en
- swissinfo.ch. Swiss government wants to adopt Council of Europe AI conventions. 12 February 2025. https://swissinfo.ch/eng/swiss-politics/federal-council-wants-to-adopt-ai-conventions-of-the-council-of-europe/88864836
- EY Switzerland. Artificial intelligence widely established in Swiss companies, but many are still in the early stages of scaling. 27 May 2026. https://www.ey.com/en_ch/newsroom/2026/05/artificial-intelligence-widely-established-in-swiss-companies-but-many-are-still-in-the-early-stages-of-scaling
- PwC Switzerland. Hopes and Fears 2026: Switzerland. 29 September 2026. https://www.pwc.ch/en/insights/transformation/global-workforce-hopes-and-fears-survey-2026-switzerland.html
- Dell’Acqua, F. et al. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality. Organization Science, 2025 (working paper first released 18 September 2023). https://pubsonline.informs.org/doi/full/10.1287/orsc.2025.21838?af=R
- World Economic Forum. The Future of Jobs Report 2025: Digest. 7 January 2025. https://weforum.org/publications/the-future-of-jobs-report-2025/digest/
- Edward Galle. European Union Artificial Intelligence Rules for Hiring Software Reach Swiss Employers Recruiting in Europe. 2 September 2026. https://edwardgalle.com/eu-ai-act-recruitment-swiss-employers-playbook-2027/
- Edward Galle. Swiss SMEs That Use AI Mostly Lack Rules on What Staff May Enter. 16 September 2026. https://edwardgalle.com/ai-swiss-smes-data-rules-skills/
- Edward Galle. One Swiss Marketing Job Ad in Eight Now Asks for AI Skills. 7 October 2026. https://edwardgalle.com/ai-skills-swiss-marketing-job-ads-brand-tech/
- Edward Galle. Human Resources Is Now the Fastest-Growing Field for AI Skills in Swiss Job Ads. 14 October 2026. https://edwardgalle.com/ai-skills-hr-jobs-switzerland/
- Edward Galle. Every Agent in Your Company Needs a Named Human Principal. That Makes It an HR Problem. 12 November 2026. https://edwardgalle.com/ai-agent-accountability-named-human-principal-hr/
- Edward Galle. Kainjoo Partners with Harvard Business Review to Power Edward Galle Training. 24 June 2026. https://edwardgalle.com/kainjoo-harvard-business-review-partnership-edward-galle-training/