""

Learning to read AI: what the AI Act says about literacy

"AI generates, a human reviews": that's the principle we used in our previous article on the AI Act to describe how the regulation found its way into our design process. But you can only review something if you know what to look for, and that's exactly what the AI Act addresses when it talks about AI literacy: the skills people need when they work with AI every day.

At Nephila, we've gradually brought AI into both our development and our management processes to support the work our people do. But AI has a personality of its own: it sounds convincing even when it's wrong, it can give a different answer to the same question every time, and it rarely admits it doesn't know. With article 4, "AI literacy", the AI Act puts the spotlight on the need to understand these quirks and know how to handle them. The regulation recognises something so obvious it risked being overlooked: behind every AI system there are people who build it and people who use it at work. Those people need to know what they're doing.

AI literacy: back to school?

In a sense, yes, though nobody's being sent back to basics or asked to become a neural network expert. The point is to know the technology you're using: what it does well and where it can go wrong. The regulation asks providers (those who develop AI systems) and deployers (those who use them) to make sure their staff have sufficient "skills, knowledge and understanding" to "gain awareness about the opportunities and risks of AI and possible harm it can cause".

The obligation covers anyone who works with AI systems on the company's behalf, including contractors and service providers, and being "non-technical" is no exemption: even people who only use AI to write or translate text need to know its specific risks, such as hallucinations. That said, there's no one-size-fits-all course. Training needs to be tailored to the knowledge and experience of the people receiving it, to the context of use, to the people the system affects and, as the European Commission suggests in its Q&A on AI literacy, to the system's risk level. Someone asking AI to draft an email needs to know different things from someone building an agent that reviews code.

What changed in 2026

This year, among other changes, the Digital Omnibus on AI (Regulation (EU) 2026/1744) rewrote Article 4. It now asks organisations to take measures to support the development of AI literacy, without requiring them to guarantee a specific level for each person. The bar is lower, but the obligation to act remains.

In Italy, the Ministry of Labour and Social Policies had already addressed the issue in its "Guidelines on AI in the workplace" (Ministerial Decree No. 180/2025), encouraging companies to invest in upskilling their staff. With Legislative Decree No. 179 of 7 October 2026, some of those recommendations have become legal requirements: "employers shall ensure that workers receive adequate information and training on the specific risks associated with the use of artificial intelligence systems, as well as on the prevention and protection measures adopted" (our translation).

The EU obligation has applied since 2 February 2025. Article 4 doesn’t provide for administrative fines, but Article 99 does for high-risk systems. The human oversight these systems require (Article 14) must be assigned to people with the necessary competence, training and authority (Article 26), and failing to demonstrate that competence can lead to heavy fines.

But fines aren't the real reason to care about this. EU lawmakers say as much themselves: AI literacy"should be a strategic priority, regardless of regulatory obligations and possible sanctions". Knowing what a powerful, fast-changing tool can and can't do helps people use it better, and leaves them with skills that last, whatever tools they use next.

How to show you're compliant

According to the European Commission's Q&A, all it takes to show you're an AI-literate company is a current internal record documenting which training and information initiatives have been put in place, for whom and when, and how and by whom they're updated. Since there's no required level of competence to reach, there's no need to measure how much people have learned either: the documentation is what proves compliance.

What we did at Nephila

What the law asks for is already at the core of our training program. Some initiatives are actual courses, while others concern the way we work with AI day to day: the regulation talks about measures, not just courses, and AI literacy is built through everyday work too.


A common baseline for everyone

We started with an introductory course on generative AI, attended by the whole team, technical and non-technical alike. As a remote-first organisation, we looked at online courses from international providers and chose a reputable learning platform with courses taught by university lecturers. The course covered prompt engineering and prompt patterns, to get more out of language models such as ChatGPT, Claude or Gemini and use them in a more advanced, structured way: at work, for study and in everyday life. We wanted everyone to start from solid foundations and to speak the same language.


Learning paths tailored to each role

For the people working on development, we put together a more in-depth path on integrating AI-based solutions into applications and technical workflows. That's the proportionality Article 4 calls for: different skills for different uses.


Shared account configuration

The accounts we use are configured the same way, so everyone works with the same tools and settings. Given the same request, we can expect more consistent results.


Collecting feedback

A form collects the experiences, questions and problems people run into in their day-to-day use. Everyone's input helps us use AI with greater awareness and update our training initiatives, which is also one of the things we need to document.


Reference documentation

We've written internal guidelines on using AI and built a shared knowledge base, where lessons learned on each project become available to the whole team. These materials are also the basis of the documentation that shows how we've met the obligation.

Regulations set the direction, but people are the ones who make the difference, through their curiosity and their eagerness to understand and share what they've learned. For us, AI literacy is something we build together, so we can keep pace with a tool that changes constantly and will play a growing role in our internal processes.


***

Sources

***

We wrote this article with the support of AI. If you're interested in how we use these tools as part of our writing process, you can read more here.

Related posts