Europe made AI literacy a legal duty 18 months ago. Most employers still have not trained anyone

It is the broadest obligation in the entire AI Act and the one least discussed. Since 2 February 2025, any provider or deployer of artificial intelligence systems in the European Union must ensure a sufficient level of AI literacy among its staff and anyone else operating those systems on its behalf.

It does not apply only to technology firms. It applies to a logistics company using an AI tool to plan routes, to an insurer scoring risk, to a law firm drafting with a model. And it reaches beyond the payroll: contractors and anyone acting on the organisation's behalf fall inside it.

Why it has been ignored

Article 4 carries no penalty of its own. There is no fine attached to failing to train your people, and after the Digital Omnibus that entered into force on 27 July 2026 the duty is one of effort rather than result.

That combination produced predictable behaviour. Compliance teams triaged the AI Act by penalty size, and an obligation with no direct sanction went to the bottom of the list.

The exposure is real nonetheless, and it arrives by two other routes. Since August 2025, an organisation whose untrained staff cause harm to a consumer or a business partner through the use of an AI system faces civil liability. And regulators treat an absence of training as an aggravating factor when enforcing the AI Act's other breaches, where fines reach up to 7 per cent of global turnover.

The article without a fine becomes expensive the moment something else goes wrong.

The credential is not the answer either

Whatever European employers do about the duty, they are unlikely to solve it by hiring graduates. The degree has been losing its function as a signal on both sides of the Atlantic.

Only 18 per cent of job postings in the United States still list a degree requirement, and formal education requirements have fallen across 87 per cent of occupational sectors. Seventy per cent of organisations now hire on skills, and 53 per cent have dropped degree requirements from at least some roles. IBM opens roughly half of its US positions to candidates without one; Google, Apple, Netflix, Walmart, Bank of America and Accenture have done the same for many roles.

American confidence in higher education has fallen to 38 per cent, from 57 per cent in 2015, and 46 per cent of Americans expect artificial intelligence to make degrees less important still.

A training market has grown into the gap

Europe records around 120,000 annual enrolments in intensive technical training, with more than 28,000 graduates in a single year. The United Kingdom alone enrols over 25,000 students a year, and participation has risen 40 per cent since 2020. Germany, France and the UK lead the field.

The sector moves from 3.77 billion dollars in 2025 to 4.09 billion this year, on course for 6.16 billion by 2031. Employment within six months of finishing runs at around 70 per cent.

What the obligation actually asks for

The text prescribes no curriculum, no certificate and no number of hours, which is where most organisations get stuck. It asks for a level of literacy sufficient for the systems being deployed and the people deploying them: an operator running a scoring model needs a different understanding from a manager approving its output.

That vagueness cuts both ways. It leaves room to comply proportionately, and it leaves no excuse for having done nothing at all, which is where most employers still are eighteen months on.

Newcomers

Among the programmes entering this space is Aiversity de la Eficacia Competente Planetaria, opening with a twenty-one module course over six months and a commitment no established provider currently publishes: if a student completes every assignment and is not invoicing by the sixth month, they continue at no cost until they are.

Its syllabus is unusual in reaching past technical ground into client acquisition, building a company, and deciding what to build at all.