AI literacy after the 2026 Article 4 change: how to document what your organisation did

Start with who uses which AI system and for what

Do not begin by buying one generic AI course for the whole company. First identify the AI systems in use and the roles that interact with them. A person using an assistant to draft internal notes faces a different context from someone relying on model output in customer decisions, and both differ from a team developing an AI-enabled product.

For each group, describe the use context: what data the system can access, whether its output affects customers or employees, where human review is mandatory, which mistakes are plausible and which internal rules already apply. This becomes the basis for choosing the learning content. Without that context, an “AI training” course can be broadly correct while missing the way people actually work.

• AI system or system class used by the role;
• purpose of use and types of data involved;
• decisions influenced by AI output;
• mandatory human review;
• common model limitations and failure modes;
• escalation rules and prohibited uses.

There is no single mandatory test or certificate

Article 4 requires measures that support the development of AI literacy. Following the 2026 amendment, it expressly states that a provider or deployer is not required to guarantee a specific level of AI literacy for an individual. The European Commission also explains that Article 4 does not create one mandatory method for measuring employee knowledge.

That does not make assessment useless. A short test, scenario exercise or acknowledgement can be a sensible part of a programme. The method should follow the risk and purpose rather than an assumption that a test score “solves Article 4”. For higher-risk roles, a stronger record may combine learning material, a scenario exercise and an accountable person’s decision.

Record the logic of the programme, not only attendance

If the organisation later needs to explain what it did, an attendance list answers only one question. A stronger record explains why a group received specific material, which version applied, who approved it and what happened after the assignment.

A minimum trace can include the group and role, AI-use context, source policy or rule set, material version, approver, assignment date, completion date, test result where a test is justified, and the action taken after an unsatisfactory result. When the AI system or its use changes, the organisation should decide whether the existing material still matches the risk.

What CrewShift can document — and what that does not prove

CrewShift can link approved material to a specific source and version, assign it to named people, record completion and results, and preserve change history. That provides evidence that a defined organisational measure took place.

It is not automatic proof of compliance with Article 4. Compliance depends on the broader context: the AI systems used, user groups, risks, programme content and the organisation’s actual governance of AI use. CrewShift does not replace legal analysis or the accountable person’s judgement about the AI literacy programme.