In Luxembourg, 59% of the companies surveyed by FEDIL, the Luxembourg AI Factory and Luxinnovation have deployed generative AI company-wide or use it in production, according to the survey AI and GenAI in Luxembourg Industry: From pilots to scaled deployment published on 2 June 2026 (136 respondents). Lack of expertise remains the main obstacle to scaling, and 45% to 48% of the companies reporting a shortage of machine learning and data analytics skills do not know the relevant training providers. Since 2 February 2025, Article 4 of the AI Act has required any organisation using AI systems to take measures to develop its teams' AI literacy; its rewriting by the regulation of 8 July 2026 states that no individual level has to be guaranteed and keeps the obligation. Here are Luxembourg's 2026 figures, what the text requires and the schemes available for training teams in AI.
The facts
- The FEDIL, AI Factory and Luxinnovation survey ran online from 27 January to 13 March 2026 with 136 respondents (20% more than in 2025), 61% of them from top management; more than half of the sites have fewer than 50 employees. 88% expect productivity gains from AI, 65% process optimisation and 64% cost savings; 70% embed AI in a digital transformation programme and 76% have implemented or plan generative AI use cases. 48% have an AI governance policy, 46% do not. The report warns that comparisons with 2025 should be read with caution, the sample being more diverse (Luxinnovation summary).
- Regulation (EU) 2026/1744 of 8 July 2026, known as the Digital Omnibus on AI, published in the Official Journal on 24 July and in force since 27 July (Parliament position of 16 June, Council decision of 29 June), replaces Article 4 of the AI Act: providers and deployers of AI systems take measures to support the development of AI literacy among their staff and the persons operating or using those systems on their behalf, taking into account their knowledge, experience, education and training and the context of use; the obligation does not require guaranteeing any specific level for anyone. A new paragraph 2 tasks the Commission and the Member States with supporting those efforts, in particular those of SMEs, with practical examples published by the Commission; a paragraph 3 asks the European AI Board for recommendations and common objectives. According to the Commission Q&A on AI literacy (updated on 20 October 2026), the article has applied since 2 February 2025, its supervision by national authorities since 2 August 2026, and no certificate is required.
- On 20 October 2026, the Ministry of Labour and the Department of Media, Connectivity and Digital Policy held with IMS Luxembourg the conference Skills in the AI Era: Preparing People and Organisations, according to the government press release. The first results of the OECD's 2026 survey presented there describe AI as widely used in companies, but with its integration into working practices often limited; in Luxembourg, 45% of companies that adopted AI recruited over the past twelve months and 14% report a reduction in headcount. LISER, in a policy brief of 10 June 2026 based on 75 million job vacancies posted between 2018 and 2023 in Belgium, France, Germany and Luxembourg, ranks the Grand Duchy among the countries with the strongest demand for AI-related skills: according to Paperjam, 78% of Luxembourg vacancies in ICT and 62% in finance included at least one AI, data or prediction skill in 2023.
Generative AI has become an everyday work tool in Luxembourg companies
Generative AI is by far the most adopted AI technology in the survey: 59% of respondents have deployed it company-wide or have use cases in production, and only 7% have no plans within twelve months. The report describes it as a standard productivity tool, which employees bring to the office as they use it at home, across all company sizes and job profiles. Starting in IT functions, it is spreading to marketing and communication, human resources, finance and customer-facing teams: document drafting and summarisation, translation, customer support, idea generation, marketing content creation.
The table below picks out the survey indicators that matter for a training decision. The results cover 136 companies, more than half of them with fewer than 50 employees, and the report asks that gaps with 2025 be read with caution.
| Indicator | 2026 result | What we take from it |
|---|---|---|
| Generative AI deployed company-wide or in production | 59% | The tool is already in the teams' hands; training covers uses that are under way |
| No generative AI project within twelve months | 7% | Companies that will stay away from it are a minority, small structures included |
| First expected benefit: productivity and efficiency | 88% | The gain depends on how each team uses the tool, hence on its skills |
| AI governance policy in place | 48% | One company in two lets its teams use AI without written rules |
| Companies short of machine learning and data analytics skills that do not know the training providers | 45% to 48% | The need exists, the offer remains invisible to nearly half of the companies concerned |
| Awareness of public support schemes, companies with fewer than 100 employees | 26% (40% above 100 employees) | SMEs are missing the existing support |
| AI maturity score, out of 5, of companies aware of public support, versus the others | 2.3 versus 1.69 | Knowing the ecosystem goes with more advanced adoption, a correlation the report calls significant |
The main barrier is the skills gap, and the labour market already rewards those skills
The report is clear on the ranking of obstacles: interest and awareness are no longer missing, expertise is. The most sought-after skills remain data analytics, AI engineering and change management. Cost became a leading barrier in 2026, because companies moving to industrial scale discover the full cost of AI, specialised skills included. An encouraging signal comes with it: the share of companies reporting a shortage is falling in almost every category, which the authors attribute to learning effects and internal capacity building. Among companies whose expectations were only partly met, the need for further training and change management is one of the reasons cited.
The training offer, for its part, remains poorly identified. Between 45% and 48% of companies reporting a shortage in machine learning and data analytics say they do not know the relevant training providers. Awareness of public support grows mainly in companies with more than 100 employees (from 17% to 40% between the two editions) and much less below that threshold (from 21% to 26%). An INFPC study published on 1 October 2025 gave the employees' point of view, on 2024 data: 48% of employees in Luxembourg said they needed to develop their AI skills, 18% had taken a training course on the subject between February 2023 and May 2024, and 84% of the AI courses recorded were delivered in person.
The labour market already values these skills. LISER's policy brief (Feimi, Gathmann, Gregory and Marguerit, 10 June 2026) shows, on 75 million job vacancies from four countries between 2018 and 2023, that AI reshapes the content of jobs more than it reduces their number: vacancies requiring AI skills rose by 29% over the period, prediction skills by 21%, judgment by 17%, data management and analysis by 16%, and vacancies in the most exposed occupations grew by about 9% a year. Employers look for staff able to integrate AI into their job, assess the quality of an output and decide, and the researchers recommend training within existing occupations, the speed at which skills become obsolete being in their view the central issue. These data stop in 2023, before generative AI became widespread; the risk they identify, a divide between employees who integrate AI and the others, has since reached every workstation.
What Article 4 of the AI Act requires, and what Luxembourg is preparing
Yes, a company that uses AI systems in its business must take measures to develop its teams' AI literacy, and has had to since 2 February 2025. The original text asked to ensure, to their best extent, a sufficient level of AI literacy; the regulation of 8 July 2026 asks to take measures to support its development and states that no individual level has to be guaranteed. Recital 8 explains the choice: stringent obligations would have weighed on smaller enterprises, whereas AI literacy should remain a strategic priority regardless of any sanction. The obligation stays with providers and deployers, in other words with any organisation using an AI tool in the course of its professional activity. The Commission had proposed this revision on 19 November 2025; the adopted text adds Member State support for SMEs and practical examples of compliance.
Key points. The obligation has been in force since 2 February 2025. Since 27 July 2026, it is an obligation of means, adapted to people's knowledge and to the context of use. No certificate, no format and no number of hours are imposed. The Commission's Q&A states that an organisation may keep an internal record of its training and initiatives, which we recommend for every action. Supervision falls to national market surveillance authorities since 2 August 2026, and penalties follow each State's own law.
In Luxembourg, that supervisory architecture is not finalised. Bill 8476 gives the CNPD the role of default market surveillance authority, coordinator of sector regulators and public contact point, according to the press release of 1 September 2026 announcing the appointment of Philippe Ney as CNPD president from 1 November 2026. In its opinion of 10 July 2026, the Council of State raised several formal objections: no allocation of powers between ILNAS and the Government Commissariat for Data Sovereignty, no rules when several complaints target the same AI system, the CSSF's scope to be clarified. It also recalls that Article 62 of the AI Act requires Member States to take measures for providers and deployers, SMEs in particular, which may consist of awareness, training, communication or practical support actions, and whose concrete implementation will have to be ensured. According to MLex (26 October 2026), the outgoing CNPD president, Tine A. Larsen, said a revised bill was being prepared for more consistent enforcement across authorities. Waiting for that text exempts no one: the European obligation already applies, and the measures taken today, documented, are the ones an authority will look at tomorrow.
The schemes to know in Luxembourg, from general literacy to specialist profiles
Public authorities have made training their main answer. In a parliamentary answer published on 3 September 2026, reported by Paperjam, the government favours adapting skills over overhauling the social model; the Accelerating Digital Sovereignty 2030 strategy makes talent development one of its six levers, with the AI4Skills project, and the Ministry of Labour has commissioned a sector study on skills needs. The same answer takes up STATEC's 2025 estimates: 90% of jobs could evolve with AI, 14% could be replaced and AI could support 55% of employees in their work, specifying that these figures are not a forecast of job losses. The schemes below exist as of 3 November 2026; we detail neither their amounts nor their eligibility conditions.
Elements of AI Luxembourg. A free, certifying online course with no prerequisites, run by the SMC with the Digital Learning Hub and the national education training institute. Since 2021, more than 7,600 people have completed it, about 1.1% of the working-age population, beyond the national target set for 2030, according to the press release of 21 May 2026; data from MinnaLearn and the University of Helsinki place Luxembourg second in Europe, behind Finland, for the diffusion of AI knowledge. This is the general literacy level, useful for a first acculturation; it does not replace training on the company's tools and use cases.
Digital Learning Hub academies. Three academies (Data & AI, Cybersecurity, Quantum Computing) inaugurated in Esch-Belval on 16 June 2026 by minister Claude Meisch, with courses started in April 2026: 81 participants selected from 913 applications, including 30 jobseekers registered with ADEM, according to ADEM; 34 attend the Data & AI academy (Python, statistics, data visualisation, machine learning) according to Chronicle.lu. Next session in April 2027, registration open from November 2026 on dlh.lu/academy. This is the specialist level.
PIX and NTAC. Since 2024, the Ministry of Labour and the SMC have worked with the ABBL on assessing employees' digital skills with PIX, a self-assessment tool; the NTAC project (New Tech Adaptation for Companies), launched in 2025 with IMS Luxembourg, adds coaching and mentoring, followed by 11 organisations. First results: a good level in data handling and collaboration, progress to be made in information searching, responsible use of digital tools and document sharing. The project continues in 2027 and 2028; interested organisations contact IMS Luxembourg.
Luxembourg AI Factory and Luxinnovation. A one-stop shop launched in March 2025, which has supported more than 150 companies, mostly SMEs, according to the review of 30 April 2026, with a training catalogue. In the FEDIL survey, Fit 4 AI is the most spontaneously cited scheme (50% of the 40 answers to an open question, a visibility indicator on a small number of answers).
Training aid and Skills-Plang. State aid for continuing vocational training in companies, which we detailed in our article on approval and co-funding, applies to training taken by employees; the law of 19 June 2025 created the Skills-Plang, financial aid for employee upskilling and reskilling run by the Ministry of Labour and ADEM, according to Paperjam. Conditions are to be checked with INFPC and ADEM; we do not claim eligibility for any training, ours included, without that check.
These schemes lay the foundation. The training that creates value in a company starts from its tools, its documents and its use cases, with written rules of use: the gaps noted by NTAC (responsible use, document sharing) and the gap between acceptance and governance measured by FEDIL each say so in their own way.
Structuring an AI training approach: seven steps
The steps below are our own practice with the companies we train; they apply to an SME as much as to a large account.
- Map actual uses. Which AI tools do your teams already use, with or without approval? The FEDIL survey shows that blocking has disappeared; the inventory, for its part, is often missing.
- Write a usage policy. Permitted and prohibited data, approved tools, human review of outputs, retention of exchanges. One page is enough to start, and it is the first measure within the meaning of Article 4.
- Segment by role. Leadership (strategy, risks, AI Act), managers (steering, change management), business functions (marketing, communication, HR, finance, customer service: concrete cases and how to phrase requests), technical profiles.
- Train on the company's own use cases. Exercises are prepared on the participants' real files; a generic demonstration changes no practice.
- Document. Content, participants, dates, materials, rules circulated: this is the internal record the Commission's Q&A mentions, and the trace of your measures.
- Measure and repeat. Adoption, reported time saved, quality of deliverables; an update at least once a year, as tools change faster than training cycles, as LISER notes.
- Mobilise the ecosystem. Elements of AI for acculturation, NTAC for support, Luxinnovation for aid, INFPC for co-funding.
A company that has taken these seven steps holds both its compliance and its return on investment: the productivity expected by 88% of FEDIL's respondents depends on what each person knows how to do with the tool.
AIxH's view
We train teams in generative AI in person, in Differdange or at the client's premises, within our approved training organisation, with more than 450 managers and executives trained. What we observe matches the survey: the tool is already there, often used without rules, and demand is for concrete cases, department by department. Our tailor-made training courses start from the participants' real files; when the company wants to go further, our AI audit and integration service configures assistants in the tools it already uses, as for the Fondation Cancer Luxembourg. Our line has not changed since the beginning: AI produces, people steer. Training teams is what keeps judgment and decisions on the human side, which is what employers look for according to LISER, and what Article 4 now asks of every company.
