The short answer
AI literacy is the knowledge and skill to use AI with judgement: knowing what a system can do, where it gets things wrong, and what those mistakes cost. Article 4 of the EU AI Act requires every organisation that builds or uses AI to work on it with its staff. It has applied since 2 February 2025, to everyone, including a small business whose only AI is ChatGPT or Copilot.
In July 2026 the Digital Omnibus rewrote Article 4. Organisations now have to take measures that support the development of AI literacy among their staff. The old promise of a sufficient level for every employee is gone. The duty is lighter, and it still applies.
What Article 4 asks of you
The AI Act distinguishes providers, who build an AI system or put it on the market, from deployers, who use one in their work. Article 4 covers both. Whether you build your own AI agent or run customer service through a chatbot, it applies to you.
The law prescribes no course, exam or certificate. It asks you to take three things into account:
Who uses the AI
The technical knowledge, experience and training of the people working with the system. A developer needs different knowledge than someone in accounts payable.
What it is used for
The context the AI runs in. Summarising internal notes asks less of people than a system that screens job applicants or advises customers.
Who it affects
The people or groups on the receiving end of the output. The more it affects customers, applicants or citizens, the more your team needs to understand what happens inside.
So the law treats AI literacy as tailored work. One standard course that everyone sits through in the same form is a weak answer to Article 4.
What the Digital Omnibus changed
The Digital Omnibus on AI is an EU regulation that simplifies parts of the AI Act and delays others. It was published on 24 July 2026 and took effect a few days later. For Article 4 it changed one sentence, and that sentence matters.
| Original text | After the Omnibus | |
|---|---|---|
| What you must do | Take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff. | Take measures to support the development of AI literacy of their staff. |
| What it hinges on | An outcome: every employee at a sufficient level. | An effort: visible, role-appropriate work on knowledge. |
| Who it covers | All providers and deployers of AI systems. | Unchanged: all providers and deployers of AI systems. |
| Since when | 2 February 2025. | Unchanged: the duty never paused. |
The shift reads like legal fine print, but it is practical. You no longer have to prove that each person reached a given level. You do have to show that you work on it seriously and in proportion to how AI is used.
After the Omnibus, Article 4 does not ask for a diploma per employee. It asks for an organisation that can show its people know what they are doing when they use AI.
The Omnibus also pushed back the high-risk rules: to December 2027 for standalone systems and to August 2028 for AI built into regulated products. That delay does not touch Article 4. AI literacy was already required, and it stays required.
Enforcement and fines
Article 4 carries no fine of its own. The AI Act does not list it among the breaches with a set penalty. That does not make it optional. National market surveillance authorities can enforce from August 2026.
What matters more is what happens when something goes wrong. If an employee acts on AI output that turns out to be false, and your organisation did nothing about knowledge or rules of use, that counts when a regulator or court looks at the incident. AI literacy is then the difference between a mistake and negligence.
How to handle AI literacy in practice
You do not need to wait for guidance or a certification scheme. These steps give you something that works and that you can show:
List the AI your organisation uses
Which tools are in use, officially and unofficially? ChatGPT, Copilot, AI inside your CRM, a chatbot on the website, an in-house agent. Ask around: most AI use lives in browser tabs, not on a software inventory.
Tie each tool to roles and risk
Write down who uses each tool and for what. A tool that rewrites emails asks less than one that processes customer data or prepares decisions.
Give each role the knowledge it needs
Everyone should know that a model can be confidently wrong and that confidential data does not go into a public tool. People who act on AI output need to know how to check it. People who build AI need to know how to test and monitor it.
Agree on rules of use
A short AI policy: which tools are allowed, what data may and may not go in, and who answers for what happens with the output. One page people actually read beats twenty nobody opens.
Record what you did
Which sessions took place, who attended and which rules apply. After the Omnibus, that record is exactly what a regulator will want to see.
Revisit it when things change
A new tool, a new role or a new use case raises new questions. AI literacy is not a one-off training, it grows with how you use AI.
What AI literacy means for a team that builds with AI
We build software with AI and build AI applications for clients. For a development team, AI literacy is more concrete than for the rest of the organisation: it means knowing where models are weak and shaping the work around that.
A developer who accepts AI-written code without reading it is not AI literate, however fast they ship. They need to spot when a model invents functions or packages, when error handling hides failures, and when a change is too large to review. We describe what that looks like in code in AI slop: what it is and how to spot it.
If you build AI applications yourself, such as an agent that handles email or reads documents, the bar moves again. The team also needs to know how to test such a system, where a human stays in the loop, and how to notice when it drifts in production. How to design that human oversight is the topic of UX for AI: keeping humans in control.
For a development team, AI literacy is not a course about AI. It is a way of working in which every line that ships has been read by someone.
Conclusion: lighter, not less important
Since February 2025, Article 4 of the AI Act has required every organisation that uses AI to work on its staff’s AI literacy. The Digital Omnibus softened that in 2026, from a guaranteed level to demonstrable effort that fits who uses AI and for what. List your AI use, give each role the knowledge it needs, agree on rules and keep a record. That covers the law, and more to the point, it keeps your team from using AI in ways it later regrets.
This article is a practical explanation, not legal advice. For questions about your own situation, talk to a lawyer who knows the AI Act.
What is AI literacy?
Is AI literacy mandatory?
Does Article 4 apply to small businesses?
What did the Digital Omnibus change?
Is there a fine for ignoring AI literacy?
Do you need an AI literacy certificate?
How do you show that you work on AI literacy?
Want your team AI literate?
In a tailored workshop your team learns what AI can do, where it fails and how to use it responsibly, with the tools you actually use.