The EU AI Act Does Not Require AI Training

A lot of AI training is currently sold on the claim that the EU AI Act requires it. It does not. Article 4 of the AI Act creates a real, enforceable obligation, and getting its shape right matters more than overstating it, because a qualified audience can check the text in five minutes.

What Article 4 says

Article 4 requires providers and deployers of AI systems to take measures to support the development of AI literacy among their staff and other people dealing with the operation and use of AI systems on their behalf. The measures are to take account of those people's technical knowledge, experience, education and training, the context the systems are used in, and the persons or groups the systems are used on.

Then comes the sentence that settles the argument, and it is new. Regulation (EU) 2026/1744, the Digital Omnibus on AI, amended all three paragraphs of Article 4 and added to paragraph 1: the obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual. That wording did not exist in the original 2024 text. The legislature went out of its way to say that no particular standard of individual competence is being imposed.

The European Commission is equally direct in its own questions and answers on AI literacy, last updated in October 2026. No strict requirements or mandatory trainings are imposed. There is no need for a certificate. The Commission also declines to prescribe any single framework or approach.

So what is the obligation, exactly

It is an obligation of conduct on the organisation, not an obligation of result on the individual. You must take measures. The measures must be proportionate to the people, the systems and the context. You do not have to certify anyone, you do not have to run a specified curriculum, and you cannot be required to guarantee that any given employee reaches a defined level.

Three further points of precision.

First, who it binds. The duty falls on providers and deployers of AI systems. Almost no finance function is a provider; most are deployers, which is enough to engage Article 4. The Commission's own material extends the reach of the duty to people broadly under the organisation's remit, giving contractors, service providers and clients as examples. So a firm whose staff use a generative assistant on client work is squarely within it.

Second, when. Article 4 entered into application on 2 February 2025. Supervision and enforcement, however, apply from 2 August 2026. The duty has therefore been live for well over a year and enforceable for several months.

Third, the penalty position, which is the opposite of how it is usually presented. Article 99 of the AI Act lists the provisions that attract administrative fines of up to 15 million euro or 3% of worldwide turnover. Article 4 is not among them. There is no direct administrative fine tier in Article 99 for breach of the AI literacy obligation; enforcement runs through Member State rules made under Article 99(1). Claims that failing to train staff exposes an organisation to eight-figure fines are not supportable on the text.

Why the honest version is the stronger argument

If the Act does not mandate training, why would a finance function invest in it? Because three separate duties converge on the same place, and two of them are harder-edged than Article 4.

The first is Article 4 itself. There is no prescribed curriculum, which means the organisation must choose a measure and be able to show it chose sensibly. A documented, role-relevant programme with a record of who completed it is a defensible answer to a supervisor asking what measures were taken. Doing nothing, or circulating a tool's instructions for use and asking staff to read them, is weaker. The Commission cautions specifically that relying on instructions for use may be ineffective.

The second is professional ethics, and this is where the real obligation lives for accountants. The International Ethics Standards Board for Accountants revised its Code to address technology, with the independence provisions effective for periods beginning on or after 15 December 2024 and the other provisions effective from the same date. The revisions are technology-neutral by design, which is exactly why they already cover artificial intelligence. IESBA's own summary of its approach, published in June 2026, puts the resulting duties bluntly: professional accountants must understand, be able to explain, and evaluate the technology they use; relying on opaque or unexplained outputs is not acceptable; and professional responsibility cannot be delegated to a machine. It also notes that having an inquiring mind became a mandatory requirement when applying the conceptual framework.

Those are not aspirations. They are the ethical standard that member bodies adopt, and they are considerably more demanding of an individual accountant than Article 4 is.

The third is the CPD obligation itself, and the key point is that no body has added an AI requirement to it. We checked ICAEW, ACCA, IFAC, AICPA and CIMA, and IESBA. ICAEW's CPD Regulations, effective 1 November 2023, set category hours with verifiable minima and mandate exactly one subject: a minimum of one hour's ethics training. ACCA mandates no subject at all within its 40 units, as our analysis of how ACCA's 40-unit rule applies to only one of three routes sets out, and the structural differences between bodies are covered in our comparison of ACCA and CIMA CPD. IESBA has deliberately chosen not to pursue further Code revisions for now, and says its planned AI output will be practical, non-authoritative guidance.

So the accurate proposition is this. No regulator and no professional body requires AI training. What exists instead is an enforceable duty to take proportionate measures, an ethical standard that already requires you to understand and be able to explain the technology you rely on, and a CPD obligation that requires learning relevant to your role. Where AI is central to the role, AI competence is what relevance means.

A note on scale

ACCA's research published in October 2026, surveying 1,178 finance and accountancy professionals across 135 countries, found AI use cited as the leading ethical challenge by 61% of respondents, up from 32% in 2024. The challenges named were overreliance, verification difficulty, reduced human judgement, confidentiality, weak human oversight, transparency and unclear accountability. ACCA's own framing of the response is notable: finance professionals are not asking for more rules, and the fundamental principles remain the ethical anchor.

That is a better case for investing in capability than any compliance threat. The duty is to be competent and to be able to explain yourself, and that duty already exists.

What to do

Write down what measures you have taken under Article 4 and why they are proportionate to your people and systems. Keep a record of who did what. Treat the IESBA requirements as the binding standard for individual accountants rather than Article 4, because they are more demanding. And resist any supplier who tells you the Act mandates their course: the amended Article 4 says in terms that no specific level of literacy is required of any individual, and the Commission's AI literacy questions and answers says there is no need for a certificate. The dates that did change, and the ones most published timelines still get wrong, are set out in our piece on the AI Act's high-risk rules moving to December 2027.