Article 4 still requires providers and deployers to take measures supporting the development of AI literacy among people who operate or use AI on their behalf. The July 2026 amendment changed the wording: it does not require an employer to guarantee a specified or sufficient level for each individual. That distinction matters when deciding what support your team needs and what evidence to keep.
What changed in July 2026
The original provision applied from February 2025. The AI Omnibus entered into force on 27 July 2026 and amended Article 4. The Commission now describes an obligation to support development, taking account of people’s knowledge, experience, education, training and the context of use. Its updated answers explicitly distinguish this from guaranteeing a particular level or measuring every employee’s knowledge.
A claim that every employee needs a prescribed certificate, compulsory course or fixed annual training duration overstates Article 4. The Commission and Member States also have a role in supporting organisations. Choosing a commercial training course should follow a practical need, rather than a claim that buying that course itself establishes compliance.
Start with the work people actually do
List the AI uses that matter to your team. A person using an approved assistant to outline an internal document faces different questions from someone reviewing an AI recommendation about a job applicant. Consider what each person must recognise, when they should check an answer and how they can get help.
For document work, a useful session might show how missing context changes an answer and how to verify a claim against the source. For a manager approving new uses, the useful support may be a briefing on intended purpose, data access, affected people and escalation. These are examples of proportionate support, not a statutory curriculum.
Training is one option
Practical guidance, supervised use, short briefings and targeted training can contribute to a programme. Pick a format people can use in the real task. A generic attendance record will tell you less about the remaining difficulty than a discussion of the errors people are still encountering.
The separate human-oversight requirements for high-risk systems need their own assessment. The Article 4 amendment does not remove those requirements. Someone assigned oversight must have the relevant competence, support and authority under the applicable provision.
Keep a useful record
As a governance practice, record the uses covered, the support offered, who took part and what needs follow-up. Explain why the measures fit the work and revisit them when the tool or task changes. A dated record helps managers understand what has happened; Article 4 does not prescribe one particular software product or certificate.
Avoid treating completion as permanent approval. If a team starts using AI with a different set of records or for a decision affecting people, revisit the support and controls. Keeping the record beside the AI inventory makes those changes easier to notice.
How Acuity can help
Acuity can help identify the practical gaps and scope support around systems your organisation already uses. Training packs and tailored working materials are supplied privately under an agreed engagement. Public guidance helps you decide whether that work would be useful.
Related guidance: practical AI training and AI governance in practice.
Legal references checked 4 October 2026: AI Omnibus in force; Article 4 guidance; Current application dates; Article 50 transparency; High-risk classification; Irish Act and AI Office. Check the provision and any transition that applies to your particular system.