Is AI training mandatory? What article 4 of the AI Act actually requires
Who exactly must be trained?
The text covers staff and other persons dealing with the operation and use of AI systems on behalf of the organisation: employees of course, but also temporary workers, integrated contractors and processors operating your systems. The measures are calibrated according to each person's technical knowledge, experience, education and training, the context of use and the persons affected by the system. In practice, three circles:
- All users of AI tools, even occasional: common foundation.
- Functions with sensitive use (HR, legal, finance, customer relations, development): business module.
- Decision-makers: governance, responsibilities, arbitrations: management module, which meets the training obligation for management bodies set out in NIS2 for regulated entities.
What should the training contain?
The notion of 'AI literacy' (Article 3, point 56) combines skills, knowledge and understanding enabling informed use and awareness of opportunities, risks and potential harms. An effective common foundation takes two to three hours:
- Understand: what an AI system is, what it does well, its structural limitations (hallucinations, biases, data freshness).
- Protect: personal data, confidential data and trade secrets in prompts; the rules of the internal charter (see ChatGPT in the enterprise).
- Control: human review, source verification, prohibition of unvalidated automated individual decisions.
- Alert: recognise high-risk use (Annex III), a prohibited practice (Article 5), an incident, and know to whom to report it.
Job-specific modules add concrete use cases for the role: a recruiter learns why CV screening will be high-risk from late 2027 onwards and what the GDPR already requires; a developer, the supply chain obligations.
What evidence to provide, and what are the risks?
No diploma or approved body is required: e-learning, workshops, internal programmes are all suitable, provided they are adapted and tracked. The typical evidence file: dated programme, record of participants with status (completed, to be renewed), validation quiz, refresh rule (annual, and at each major tool change), integration into the onboarding process for new arrivals.
Article 4 is not subject to a specific fine in the grid of Article 99, but its non-performance weighs everywhere else: it aggravates the assessment of any breach of the AI Act, undermines the company in litigation (employee, customer) involving wrongful use of AI, and contradicts GDPR accountability when AI processes personal data. Conversely, a tracked programme is the first proof of diligence produced in the event of a problem.
Pool with GDPR and NIS2: a single pathway
Awareness-raising obligations pile up: data protection (GDPR accountability), cyber hygiene and training (Article 21 of NIS2), AI literacy (Article 4). Handling them in three separate pathways exhausts teams; a single, modular compliance pathway covers them together: annual common foundation (data, cyber, AI), function-specific modules, executive module, with a single evidence record. This is the approach of our eLearning integrated into the platform: assignment by role, quiz, certificates and coverage dashboard, with evidence building itself automatically.
Implement the system in 30 days
- Week 1: identify users and uses (the shadow AI mapping provides the list).
- Week 2: finalise the programme (foundation + modules) and the refresh rule; validate the AI charter that serves as support.
- Weeks 3 and 4: deploy the common foundation to all identified users, track, then schedule the job-specific modules.
- Include training coverage (percentage of users trained) in the compliance dashboard presented to management.
Take action
Manage your compliance with Data Comply One: all-in-one software and outsourced DPO/CISO experts.
Discover the AI Act software →Test my AI compliance →Sources
- Regulation (EU) 2024/1689 (AI Act), Articles 3 (point 56), 4, 5 and 99, and Regulation (EU) 2026/1744 (AI Omnibus) amending Article 4, EUR-Lex.
- European Commission, AI Office: Q&A on AI literacy and collection of practices.
- Directive (EU) 2022/2555 (NIS2), Articles 20 and 21 (training and cyber hygiene), EUR-Lex.