The EU wants workers to be AI-literate. But it has lagerly left companies to decide what that means and how to prove it.
The EU AI Act requires companies that develop AI systems or use them in their work to take measures to support AI literacy among staff and others working with AI on their behalf. That means the obligation is not limited to companies developing AI. It can also apply to ordinary businesses using AI tools as part of their work.
But the rules don’t tell companies exactly what AI literacy should look like, or how to demonstrate that workers have achieved it.
“Now, no level (of AI literacy) needs to be ensured, employers only need to show that they made an effort.”
— Siddhi Pal, co-author of Interface’s AI literacy policy brief
“Currently, companies are assessing AI literacy at an individual firm level. This means there’s a lot of variation, with many focusing on AI adoption and productivity to ensure staff can use AI tools well, and maybe less focused on ethics,” Catherine Schneider from tech policy think-tank Interface, who co-penned a policy brief on the issue, told EU Perspectives.
“What we’re left with is little to no understanding of what kinds of trainings are actually effective for AI literacy, and absolutely zero consistency for workers to demonstrate and transfer their knowledge across different jobs and companies.”
You might be interested
AI Omnibus weakens literacy obligation
The AI Omnibus softened the obligation. It replaced the need to ensure a “sufficient level of AI literacy” with the requirement of “measures to support the development” of AI literacy.
Siddhi Pal, co-author of the Interface brief, told EU Perspectives the change shifts the emphasis away from achieving a particular level of competence.
“The real risk is that this becomes a box-ticking exercise.”
— Siddhi Pal, co-author of Interface’s AI literacy policy brief
“In practical terms, the obligation has moved from a result to an effort. Though ambiguous, the requirement was for employers to ensure a ‘sufficient level of AI literacy’. Now, no level needs to be ensured, employers only need to show that they made an effort.”
The change does not remove the stronger requirements applying to high-risk AI systems. Employees working with such systems still have to receive training, while the systems must remain subject to human oversight.
Risk of a ‘box-ticking exercise’
Interface argues that policymakers should develop a validated way of measuring whether training actually improves workers’ understanding. The goal is to avoid employers completing a training course without demonstrating capabilities of recognising and responding to AI-related risks.
“The real risk is that this becomes a box-ticking exercise,” Pal said. “If all you have to do is make an effort, and there’s no level anyone has to reach and nothing you have to measure, then a training log isn’t weak proof that you’ve complied. It’s full proof.”
Instead, Interface suggests that different uses of AI should require different levels of knowledge. Its own proposal divides AI literacy into three cumulative tiers. It’s based on how closely a worker interacts with AI and the risks associated with that interaction instead of seniority, job title or technical background.
Uneven AI adoption makes standardisation harder
According to Eurostat, 20 per cent of EU enterprises with at least 10 employees used AI in 2025. The share rose from 13.5 per cent in 2024. At the same time, nearly a quarter of European workers use AI at work. But adoption varies between member states. At the extremes, it is Denmark with 42 per cent, compared with just 5.2 per cent in Romania.
That uneven adoption is mirrored by differences in how companies approach AI literacy, Interface argues.
“The result is a fragmented ecosystem of employers relying on a variety of metrics for capturing AI usage and literacy, like general-purpose learning programs and certifications or systems monitoring like employee token usage”, said Schneider.
