AI literacy at work.
Beyond the prompt.

Know the system. Question the answer.

AI literacy is the ability to understand, use, evaluate, and make informed decisions about AI. For a team, that means understanding the tools in its actual workflow: what data they receive, what they produce, what they can access, and what a person is responsible for checking.

Prompt writing is one small skill. People also need to notice when an answer is unsupported, recognize sensitive information, check source material, and understand whether an AI output can influence a decision about someone else. The right depth depends on the task and the system.

This practical view is consistent with research that treats AI literacy as a set of competencies, including critically evaluating AI and communicating or collaborating with it—not simply operating an interface.

Article 4 asks organisations to support learning.

For providers and deployers of AI systems, Article 4 requires measures to support the AI literacy of staff and other people who operate or use systems on their behalf. The European Commission says those measures should take account of people’s technical knowledge, experience, education and training, as well as the system’s context and the people affected by it.

Following the 2026 amendment, Article 4 does not prescribe a single training format or require organisations to guarantee a specific level of AI literacy for every individual. The Commission’s guidance also says Article 4 does not require an individual knowledge test, a certificate, or a particular governance structure. It still requires providers and deployers to take measures; the appropriate response depends on the role, system, risks, and use.

Article 4 has applied since 2 February 2025, and supervision and enforcement began on 2 August 2026. Separate duties remain relevant: for example, deployers of high-risk AI systems must ensure that staff have the training needed for human oversight. Article 4 is one part of the AI Act, not a substitute for checking the obligations that apply to a particular system.

Start with a task, not a slide deck.

1. Choose one real workflow. Name the AI system, its purpose, and the people who use or rely on its output.
2. Trace the inputs and access. Decide what information may be entered, what the system can retrieve, and which tools or accounts it can reach.
3. Compare a routine case with a difficult one. Include missing context, a confident mistake, sensitive data, or an instruction hidden in source material.
4. Practice verification. Identify what must be checked against an authoritative source, who reviews it, and what happens when the evidence is incomplete.
5. Set the human boundary. Make clear which actions need approval, who owns the final decision, and how to stop or report a problem.
6. Keep a short record. Note the system and workflow covered, the guidance shared, and when the material should be revisited.

A short exercise using a team’s real tools can reveal gaps that a generic introduction misses. Revisit it when a system, workflow, or risk changes.

Teach the decisions each person owns.

These are learning prompts, not a universal compliance checklist. Map them to the organisation’s role as provider or deployer and to the specific systems and contexts it works with.

Make literacy part of system design.

Read why fluent output needs evidence ↗, how to evaluate an agent ↗, and where agent permissions belong ↗. For school and self-study, see what the AI tutoring research says ↗.