AI literacy training became a legal duty in the European Union on 2 February 2025. Article 4 of the EU AI Act requires providers and deployers of AI systems to take measures to ensure a sufficient level of AI literacy among their staff and other people operating those systems on their behalf, taking into account those people’s technical knowledge, experience, education and training, and the context in which the systems will be used. The enforcement and penalty machinery for that duty became available from 2 August 2026. The obligation is therefore old enough that a regulator asking about it will expect to see what you did, not what you plan to do.
Article 4 catches organisations out in a particular way. It binds deployers as well as providers, so buying a tool does not move the duty to the vendor, and the standard it sets is relative rather than fixed. Sufficient literacy for a marketing team drafting copy is not sufficient literacy for a team screening job applicants. A single generic session delivered to everyone identically is weak evidence, because it cannot show that context of use was taken into account.
The AI Literacy Training course
Our AI Literacy Training course is built to satisfy Article 4 as written and to leave a record behind. It runs to 15 lessons, each with a graded quiz, and issues a certificate on completion.
The lessons work from how models behave to what people must do about it. Learners cover what a large language model is doing when it produces an answer, why fluency and accuracy are unrelated, and why a model states a wrong answer with the same confidence as a right one. They cover hallucination and how to spot it in your own domain, prompt injection and why untrusted text is an attack surface, bias in training data and in deployment, and privacy exposure when information is pasted into an external service. They cover the human oversight the Act expects, meaning what it takes for a person to be genuinely able to override a system rather than nominally in the loop. They cover the categories of decision that require a person to make the call, incident reporting, and the disclosure duties under Article 50 that have applied since 2 August 2026. They also cover the Article 5 prohibitions, so that staff can recognise a use case that is not merely risky but banned.
The quizzes exist for a reason beyond pedagogy. A completion record with graded assessment shows that people were tested on the material, which is materially better evidence than an attendance list.
Why the evidence matters more than the slides
The reason we design literacy training around records is the subject of our own peer-reviewed work. Our paper on assurance laundering was accepted at the Alan Turing Institute’s Trustworthy Digital Infrastructure conference in 2026. Its argument is that assurance activities routinely produce artefacts that look like evidence while establishing nothing, because nobody checks whether the artefact supports the claim attached to it. Literacy training is a clean example. A slide deck delivered once to a room of people who were checking email is an artefact. It is not evidence that a sufficient level of AI literacy exists among the people operating your systems, and it will not read as evidence to anyone examining it later.
Who needs it
Every organisation using AI systems in the EU, or whose system output is used in the EU, needs a defensible answer on Article 4. In practice the groups needing differentiated content are general staff using assistants, managers who approve AI-supported decisions, HR and recruitment teams where employment law also applies, developers building on top of models, and procurement teams assessing vendors. The course covers the shared baseline; role-specific routes sit alongside it in the wider estate.
Frequently asked questions
What is AI literacy under the EU AI Act? Article 4 defines the outcome rather than the syllabus. It requires measures ensuring staff and people operating systems on your behalf have skills, knowledge and understanding appropriate to their role and to the context in which the system is used.
When did the AI literacy obligation start? It has applied since 2 February 2025. The enforcement and penalty provisions became available from 2 August 2026, so the duty predates the ability to sanction it by eighteen months.
Does Article 4 apply to companies that only use AI, not build it? Yes. The obligation is placed on deployers as well as providers. Procuring a system from a vendor leaves the literacy duty with the organisation whose staff operate it.
How much AI literacy training is enough? The Act sets a relative standard tied to role, prior knowledge and context of use, so there is no fixed hour count. A defensible programme differentiates content by role and keeps records showing who completed what.
Does AI literacy training need to be certified? The Act does not require certification by a third party. It requires that measures were taken and, in practice, that you can show it. Graded assessment and completion records are the practical form that proof takes.
Is AI literacy the same as AI safety training? No. Literacy is the baseline understanding Article 4 requires across an organisation. Safety training is deeper technical work on measuring and controlling system behaviour, aimed at the people building or assuring systems.
Enrol
The AI Literacy Training course covers Article 4 in 15 lessons with graded quizzes and a certificate, and can be rolled out across a whole organisation with completion records retained. If you would rather have your existing literacy position examined before you buy anything, our two-week AI assurance claim audit tests whether the evidence you hold would satisfy someone asking for it. Contact us through the site to arrange either.