Article

AI literacy training required: how to get it right

Article 4 of the EU AI Act is already in force. A generic ChatGPT course does not comply. What good AI literacy training actually contains and how to document it properly.

Written by Loek Delahaye, founder, Delahaye Solutions · ex-CTO with 10+ years in software and AIPublished:
Short answer
  • Article 4 of the EU AI Act requires companies to ensure that employees using AI tools have adequate AI literacy. This has been in force since 2 February 2025.
  • A generic course does not comply: training must be role-specific — a recruiter using AI scoring needs different knowledge than a marketer using Midjourney.
  • Documentation is mandatory: you must be able to show that training was given, who completed it, when and what the content was.
  • Three steps: (1) inventory which AI tools are used per role, (2) determine what each role minimally needs to understand, (3) train and document — and repeat when tools change.

Why a generic AI course is not enough

Many companies respond to the AI literacy obligation with a collective course: an online module covering what AI is, how ChatGPT works and what the risks are. That is better than nothing, but it does not satisfy Article 4 of the EU AI Act. The obligation is more targeted: each employee must have sufficient AI literacy for the specific AI tools they use in their role. A recruiter using AI scoring software needs different knowledge than an accountant using Copilot for Excel formulas. The Act does not specify how many hours the training should take — but it does require you to be able to demonstrate it when a supervisor asks. That sets requirements for both the content and the documentation.

What Article 4 requires

Three requirements for good AI literacy training

The EU AI Act is deliberately vague about the exact form — but three requirements can be clearly derived from the text and enforcement practice in comparable European legislation.

Requirement 1: align with the AI tools actually being used

Article 4 speaks of 'sufficient AI literacy' in the context of an employee's specific tasks. That means the training must connect to the AI systems that employee actually uses. Start with an AI inventory per team: which tools are in use, for which tasks and with what degree of AI support? Based on that, determine per role what the minimum baseline knowledge is. For a light user (someone who occasionally uses ChatGPT for a summary) a shorter training is sufficient compared to a recruiter whose selection decisions are partly based on AI output.

Requirement 2: content that employees genuinely understand

Effective AI literacy training answers three core questions for each role: how does the AI in the tool I use work (at a high level), what are its limitations and possible errors, and what do I do when the AI makes a mistake or a customer or candidate raises an objection? For high-risk roles (think recruitment AI or diagnostic AI in healthcare), add: who is the overseer who checks my AI decisions and how do I report an incident? This does not need to be extensive — but it must be concrete, not abstract.

Requirement 3: recorded documentation that withstands an audit

The AI literacy obligation is only demonstrable if you have documented it. Minimum documentation: a training record per employee (name, role, which training, date, duration and content) and a version history when you update the training as tools change. Store this as part of your AI register. When a national supervisory authority — with enforcement powers from 2 August 2026 — asks for evidence, you don't want to spend hours searching through scattered emails and presentations. A structured file per employee is the standard in comparable European legislation (GDPR, ISO 27001) and is the benchmark here too.

Step by step

How to build an AI literacy training programme

A practical plan: (1) Carry out an AI inventory — list per team or role which AI tools are used and how intensively. Two columns: tool and the AI function being used. (2) Determine the required knowledge level per role — light users versus decision-support use require different levels. For high-risk applications (Annex III: recruitment, diagnostic, credit) check the additional requirements for deployers. (3) Assemble the training content — role-specific, focused on the concrete tools, covering the three core questions above. Internal or external: the Act does not prescribe which, as long as the quality is adequate. (4) Document immediately — create a training record per employee and include it in the AI register. (5) Schedule a repeat — when you introduce new AI tools or significantly upgrade existing ones, update the training and the record. AI literacy is not a one-time checkbox but an ongoing obligation.

Frequently asked questions

Frequently asked questions about AI literacy training and the EU AI Act

Does Article 4 also apply to small businesses and freelancers?
Yes. Article 4 sets no threshold based on company size. If you use AI tools in your work as a freelancer or small business, the obligation applies to you too. The proportionality principle applies: a freelancer who occasionally uses ChatGPT is expected to produce less extensive documentation than a company with dozens of employees in high-risk roles. But the principle — understand what the AI does and document that you understand it — applies to everyone.
What counts as adequate AI literacy?
The Act does not define it precisely, but mentions 'technical knowledge, experience, education and context-specific circumstances' as factors. In practice this means: an employee must understand how the AI tool they use works (at an understandable level), what its limitations and risks are, and how to act when the AI makes an error or a customer raises an objection. For light users this is relatively concise; for employees who use AI output in decisions about people (recruitment, credit, diagnosis) the bar is higher.
Can we use an existing AI course from an external provider?
Yes, if that course connects to the specific AI tools your employees actually use. A generic course on 'what is AI?' generally does not satisfy the requirement of role-specific literacy. Check whether the provider offers custom modules for the tools your team uses, or combine a generic basic module with role-specific additions you compile yourself. In either case, document who completed the training and what the content was.
What are the consequences of not getting this right?
National supervisory authorities receive their enforcement powers from 2 August 2026. Fines for violations of the EU AI Act can reach €15 million or 3% of worldwide annual turnover, whichever is higher. Article 4 is a baseline obligation; non-compliance can also weigh in the assessment of broader compliance, particularly for companies using high-risk AI. Starting with a structured approach now is far cheaper than fixing it after an enforcement action.
How quickly can we comply with Article 4?
With Delahaye Solutions' AI Act Scan you can be in order in about a week: we inventory your AI usage per role, assemble role-specific AI literacy training, deliver the training to your team and provide an audit-ready training record as part of the full AI register. Fixed price: €750 for small teams and freelancers; €1,500 for businesses with 10-50 employees.
Read more

Get your AI literacy training audit-ready?

Book a free intake for the AI Act Scan. We look together at your AI usage, assemble role-specific training and deliver a documented dossier — no commitment.

Book a free AI Act intake →

Free and without obligation · Reply within one business day · Fixed price up front, no surprises