A team leader at a secondary school is asked in the Dutch department meeting whether the grading tool two colleagues have used since the summer is allowed under the AI Act. The same week, the staff training day has an hour on AI literacy on the agenda, because that is supposed to be mandatory. The two questions belong together, and the second answer is shorter than it sounds.
Article 4 of the AI Act has applied to schools since 2 February 2025. A school that uses AI systems under its own responsibility is a deployer, and takes measures to ensure a sufficient level of AI literacy among the staff who work with them. There is no course in the text, no certificate and no number of hours. What counts as sufficient depends on knowledge, experience and the context of use, and the Act explicitly names the people the AI is applied to. In a school, those are pupils.
Why does Article 4 apply to a school?
The Act addresses providers and deployers. A school does not build AI, it uses it: a teacher having a chatbot draft a test, a form tutor having a letter to parents rewritten, a learning tool with an adaptive feature. Whoever uses an AI system under their own authority is a deployer; the legal entity behind the school is the board. The text is in Regulation (EU) 2024/1689; what Article 4 says and why there is no course in it is set out in AI literacy in 2026.
The Digital Omnibus of July 2026 turned it into an obligation of effort: take measures that support the development of AI literacy. That changes little about what you do, and something about the tone of what you record: not "our teachers are AI literate", but "this is what we did per role, and this is why it fits".
Two phrases in Article 4 carry more weight in education than elsewhere. "On their behalf" covers supply teachers, trainees and the external tutor running exam preparation. And "the persons on whom the AI systems are to be used" are minors here, with name, date of birth, family situation, care history and results in one place. The European Commission says in its questions and answers on Article 4 that context determines what is sufficient. For a school, that is the reason to look per role.
Why is the bar higher in education?
Because education is one of the eight areas in Annex III. The Act lists AI that determines admission to a programme, evaluates learning outcomes, determines the appropriate level of education, or detects prohibited behaviour during tests. An AI system intended for one of those purposes is high-risk. From 2 December 2027 the duties of Article 26 apply to it, including human oversight by people with the knowledge and the authority to intervene. It turns on the intended purpose the provider has documented, and the precise application differs per system and per school.
A chatbot that rewrites a lesson sheet or invents a practice test does not fall under that. Producing, summarising and translating text is not in Annex III. It changes once a tool is built to grade essays, place pupils at a level, or watch for irregular behaviour during a digital test. Where that line runs is covered in when is an AI application high-risk.
One category is not high-risk but prohibited: AI that infers pupils' emotions in an educational setting, except for medical or safety reasons. A tool that scores "engagement" in an online lesson gets switched off. The other obligations are in the EU AI Act for employers.
What does sufficient mean per role?
The question is not how many hours, but what someone in this role needs to understand to use AI responsibly. That differs, and the differences are what you record.
| Role | What the person needs to understand | What the school records |
|---|---|---|
| Teacher using AI for lesson material and admin | Which tools are allowed, which pupil data stays out, that an answer gets checked | The explanation this group received, and the decision per tool |
| Teacher or form tutor using AI in feedback and assessment | That a grade stays with the teacher, which tools were assessed for this, and when a tool assesses rather than assists | The agreement per tool, and who was briefed |
| Special needs coordinator, tutor with care files | That care notes and a development plan contain special category data that belongs only in an assessed environment | Which environment that is, and the briefing this group received |
| Exam officer, assessment committee | When a tool falls under Annex III, and what human oversight then involves | The assessment per system, with a date |
| ICT and procurement | Vendor, data processing agreement through the sector privacy covenant, training policy, country, risk | The questionnaire and the register of assessed tools |
| Board and school leadership | The statuses per tool, who assesses a new tool, and their own responsibility as deployer | The policy, the decisions and who took them |
The largest group sits in the first row, where the briefing is simplest. Why not everyone needs the same briefing is in does every employee need AI training.
What does this look like at a school?
A board with eight schools runs a staff training day at each of them. The message is the same everywhere: Copilot with the school account is allowed, pupil data does not go into free tools, when in doubt ask the ICT coordinator. The attendance lists go in the folder.
An inventory six months later shows that next to Copilot, the free version of ChatGPT, MagicSchool, Curipod and a grading tool for writing assignments are in use. No decision was ever taken on the last two. The Dutch department has the grading tool propose a first mark for essays, and usually adjusts it.
That is not an incident. It shows two things. The training day was one moment and the work went on, with tools that did not exist yet. And the grading tool raises a different question from the chatbot: does the tool assess, or does it help the teacher assess? The board asks the vendor which intended purpose it has documented, decides on the tool, and briefs the department: the mark stays with the teacher.
How do you show it?
With your own records, since no certificate is required. Three kinds of evidence each say something different: the briefing per role and who received it, the decisions per tool, and what happened afterwards in daily work. The last one is missing at most schools, and it is what a supervisory authority eventually asks about. How the three relate is set out in how to demonstrate AI literacy.
The third part requires that the school's decision is present when a teacher opens a tool, and that you can see afterwards what happened, without knowing who it was. BeeSensible links a catalogue of AI tools to the use observed in the organisation, shows the decision per tool in the browser at the moment someone opens that tool, and counts what the person did next. The figures are aggregated and carry no user id. What that means for a school board is on the page about education.
