AI training for staff: the EU literacy duty after 2026, and what companies do

In 2025, 70.31% of EU enterprises that considered using artificial intelligence (AI) and did not use it gave lack of relevant expertise as a reason [5]. For enterprises with 250 or more employees the share was 65.11% [5]. It is the most common reason in Eurostat's database for that survey year. (Eurostat's article text gives 70.89% "in 2025", but the database files that value under 2024, so this page uses the database.)

Staff knowledge is also a legal duty in the European Union (EU). This page explains what the EU AI Act requires of a large company's staff training as of 26 October 2026, after the 2026 changes. It then gives what the US National Institute of Standards and Technology (NIST) asks, how many companies in the United States (US), the United Kingdom (UK) and Canada train their staff, and what the research says about training that works. It is part of the guide on enterprise AI adoption.

What the EU AI Act requires now

Article 4 of the EU AI Act is the AI literacy duty. The Act defines AI literacy as the skills, knowledge and understanding that let people make "an informed deployment of AI systems" and be aware of AI's opportunities, risks and possible harm [1]. The original text, in Regulation (EU) 2024/1689, said: "Providers and deployers of AI systems shall take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf" [1]. A provider is the organisation that develops an AI system and puts it on the market. A deployer is an organisation that uses one under its own authority, as a company does when it gives staff an AI tool.

Regulation (EU) 2026/1744 replaced that text. The new Article 4 says: "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" [2]. The new text also tells them to take into account their staff's technical knowledge, experience, education and training, the context the systems are used in, and the people the systems are used on [2]. It adds: "This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual" [2].

Point	Original Article 4 (2024)	Article 4 as amended (2026)
The duty	Take measures to ensure, to their best extent, a sufficient level of AI literacy [1]	Take measures to support the development of AI literacy [2]
Level required	A sufficient level [1]	No specific level guaranteed for any individual [2]
Help from the EU	None in the article	The Commission publishes practical examples; the AI Board adopts recommendations [2]
Who is covered	Staff and other persons acting on the company's behalf [1]	The same [2]

The new Article 4(2) says "the Commission shall publish practical examples of how to comply with that obligation on the single information platform" [2]. Article 4(3) says "The Board shall adopt recommendations, taking into account European competence frameworks" [2]. The Board is the European Artificial Intelligence Board set up by the AI Act [1].

The EU gave its reasons in the regulation's recitals. Stakeholders said a strict duty to ensure a sufficient level of AI literacy would not suit every type of provider and deployer, and the regulation says data indicates that such a duty "creates an additional compliance burden", naming smaller enterprises as the firms most affected [2]. For a large company, the duty now asks for measures to be taken, with no level to reach. The duty still applies.

The dates

Regulation (EU) 2026/1744 was signed on 8 July 2026, published in the Official Journal on 24 July 2026, and entered into force "on the third day following that of its publication", which is 27 July 2026 [2]. The European Commission's question-and-answer page says it "entered into force in mid-July 2026" [3]. This page uses the regulation's own dates. Article 4 sits in Chapter I of the AI Act, and "Chapters I and II shall apply from 2 February 2025" [2]. So the duty has applied since 2 February 2025, in its new form from 27 July 2026.

Enforcement sits with each country. The Commission says: "The national market surveillance authorities will start supervising and enforcing the rules as of 2 August 2026" [3]. The Commission's page was last updated on 20 October 2026 [3].

What counts as evidence

The Commission's page answers the practical questions. On proof, it says: "There is no need for a certificate. Organisations can keep an internal record of trainings and/or other guiding initiatives" [3]. On format, it says there is no single model for AI literacy, and that "no strict requirements or mandatory trainings are imposed" [3]. It adds that "having different levels of training or learning approaches could be appropriate" [3].

It also warns that "in many cases, simply relying on the AI systems' instructions for use or asking the staff to read them might be ineffective" [3]. By that account, sending staff a vendor's manual and nothing else may fall short.

"Other persons" in Article 4 reach past employees. The Commission's example: "It could be, for example, a contractor, a service provider, a client" [3]. A large company that has contractors running AI systems on its behalf should include them in the plan.

For high-risk systems, a second duty applies. Under Article 26(2), "Deployers shall assign human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support" [1]. Regulation (EU) 2026/1744 left this text in place, and the deployer duties for Annex III high-risk systems apply from 2 December 2027 [2]. Human oversight means a person who watches the system and can take over from it or stop it. The page AI governance for a large company covers the high-risk duties in full.

What NIST asks

The NIST AI Risk Management Framework is voluntary. Item GOVERN 2.2 says: "The organization's personnel and partners receive AI risk management training to enable them to perform their duties and responsibilities consistent with related policies, procedures, and agreements" [4]. The training it asks for is tied to each person's duties and to the company's own policies, so it follows from the governance work.

What companies do

The figures below come from four official surveys, each with its own base.

Country	Measure	Figure	Base and period	Source
United States	Trained current staff to use AI	31.4%	Firms with 250 or more employees that used AI, 17 November 2025 to 8 February 2026	[6]
United States	Trained current staff to use AI	15.0%	All firms that used AI, same period	[6]
United States	Made no changes at all to use AI	64.3%	All firms that used AI, same period	[6]
Canada	Trained existing employees	68.1%	Businesses with 100 or more employees that used AI, 2026	[9]
Canada	Trained existing executives	51.7%	Same base	[9]
United Kingdom	More than half the workforce has had AI training	11%	Businesses with 10 or more employees, June 2026	[10]

The US figures come from the Census Bureau's AI supplement. Because of Executive Order 14434, signed on 29 September 2026, the Census Bureau now calls AI "Super Intelligence" or "SI" in its files [7][8]. Its questions asked about "Artificial Intelligence (AI)" until the collection that began on 19 October 2026, and the supplement was collected from November 2025 to February 2026, so this page says AI [8]. Among large firms that used AI, 48.5% made no changes at all to use it [6].

In the UK, the Office for National Statistics (ONS) reports: "Only 11% of businesses (with 10 or more employees) reported that more than half of their workforce has received AI-related training" [10]. Firms that name a lack of AI expertise as a barrier are more likely to train: 62% of them report training or retraining existing staff to bring in AI skills, against 26% of businesses that report no barriers [10].

In the US, 64.3% of firms that used AI changed nothing to use it, and 15.0% trained current staff [6]. A related figure is on the page how many companies use AI: 22.9% of respondents at large US firms did not know whether their own firm used AI.

What the research says works

Two sources link training on real tools and real work with more use of AI and more reported benefit.

Anders Humlum and Emilie Vestergaard surveyed workers in 11 occupations exposed to AI chatbots in Denmark for a working paper of the National Bureau of Economic Research (NBER), revised in March 2026. They found that "adoption and reported benefits peak in workplaces that combine encouraged use with enterprise chatbots and training" [11]. In such workplaces, 93% of workers had used AI chatbots at work, 28% used them daily, and 19% reported saving more than one hour per day [11]. An enterprise chatbot is a chat tool the employer provides and controls. The figures are self-reported and show a correlation, so they cannot separate the effect of training from the other two factors.

A report by the OECD (the Organisation for Economic Co-operation and Development), Boston Consulting Group (BCG) and INSEAD rests on a survey of 840 enterprises that use AI in the G7 countries, plus 167 in Brazil, run in 2022 and 2023. In its policy advice, the report says training should be tailored to industry or business needs, and that "Training on real-world projects, using AI systems and datasets common in specific areas of business" has special value [12]. BCG is a co-author of the report.

A training plan by group

The plan below is Reveneau's suggestion, built from the sources above. It keeps the Commission's point that levels can differ, and the NIST point that training follows each person's duties.

Group	What they need to know	Evidence to keep
Board and executives	What AI the company uses, the rules that apply, the risks named in the governance policy	Dated briefing record
Staff who use chat tools	Which tools are approved, which data may go into them, how to check an answer before using it	Course record per person
Staff in high-risk processes	How the system works, its known limits, when and how to take over or stop it	Training record plus the oversight assignment under Article 26(2)
Engineers who build AI features	Security risks such as prompt injection, data rules, how evals test the system	Course record and the eval results for their systems
Contractors and service providers	The same as staff in the same role	Clause in the contract and a record

An eval is a written, repeatable test of an AI system's answers against cases the company chooses. The guide on LLM evals explains how they are written.

Reveneau's advice is to train on the tools staff will use and on their own work, and to keep a dated record of each session. When a new tool is approved, train its users before it goes live. The page choosing your first AI use cases shows how to pick the first work, and AI security risks for a large company lists the risks engineers need to know.

Where Reveneau fits

Reveneau is an AI software development consultancy with offices in San Francisco and Vancouver. Reveneau says AI writes all of its code, every change must pass an eval suite written from the specification, and it takes responsibility for the whole project through production and after release. The eval suite for a system a company builds is also training material: it shows staff, case by case, what the system should and should not do. To talk about a project, see Reveneau's services or contact Reveneau.