AI in education: what the EU AI Act means in practice for schools and educational institutions

Updated 3 July 2026. On 29 June 2026, the Council gave its final green light to the Digital Omnibus on AI. The act still awaits publication in the Official Journal and will enter into force on the third day after publication. The revised application dates are set at 2 December 2027 for stand-alone high-risk AI systems and 2 August 2028 for high-risk AI systems embedded in regulated products. The AI literacy obligation of Article 4 does not change: it has applied since 2 February 2025.

AI has swept into the classroom at high speed: with pupils, with teachers and in the back office of schools. Admissions, assessment, proctoring, plagiarism or AI detection used in assessment, and personalised learning systems that steer learning paths: a good part of this can fall under the high-risk category of the EU AI Act. And although the strictest requirements take effect later than first planned, the basic obligation of the law already applies today: AI literacy for staff and other people who deal with AI systems on behalf of the institution.

Why education is a separate category

In public perception, AI in education is mainly a debate about "ChatGPT in homework": do pupils use it, how do you detect it, what is and is not allowed. That is an important question but it only scratches the surface of what is at stake. For the EU AI Act, the emphasis lies elsewhere. The law focuses above all on AI systems used in decisions or assessments about pupils and students: admission, allocation, assessment, guidance, progression and exam monitoring.

This is set out expressly in Annex III of the regulation. AI used for access, allocation, assessment or the monitoring of pupils during tests is in principle high-risk. In principle: the law has an exception for applications that pose no significant risk to health, safety or fundamental rights, for example because they only perform a preparatory or supporting task. That has to be assessed and documented per application. For schools and educational institutions, it means one thing regardless: the EU AI Act is not a side issue but touches the heart of the core educational processes.

What schools often do not think of

In conversations with school boards and education coordinators, the surprise comes up again and again that so many existing tools are relevant under the law. Consider:

Insofar as these tools qualify as an AI system under the law, they fall under the EU AI Act. Not every example is automatically high-risk: the classification depends on the specific function, the intended purpose and the impact on pupils or students. Especially when such systems feed into assessment, findings of fraud or sanctions, the AI Act relevance grows. A signalling function without automatic consequence calls for a different assessment than a system that effectively takes part in the decision. But the AI literacy obligation of Article 4 applies to all these cases, and that obligation rests with the institution as deployer. Exactly what Article 4 of the EU AI Act requires is something we set out in a separate insight.

Which deadlines actually apply now?

Following the Council's final green light on the Digital Omnibus of 29 June 2026, these are the dates now foreseen:

Already in force. The AI literacy obligation of Article 4 has applied since 2 February 2025, for every organisation that uses AI systems and therefore for every school. The prohibited AI practices under the law, such as emotion recognition in educational institutions (outside medical or safety purposes), also already apply.

2 December 2026. According to the agreement, this is when the new ban on AI systems for non-consensual intimate imagery and child sexual abuse material comes into force, among other things, a topic that schools, in their dealings with pupils and social media, unfortunately cannot avoid.

2 December 2027. The high-risk requirements for stand-alone Annex III systems become applicable. For education, that concerns AI in admission, assessment, progression decisions and exam monitoring: risk management, data quality, human oversight, logging, technical documentation. The conformity assessment lies with the provider of the system; the institution, as deployer, has its own obligations, such as using the system in line with the instructions, organising human oversight and monitoring how it works.

2 August 2028. AI built into products that already fall under other EU product rules follows then.

Important to add: the act still awaits publication in the Official Journal; until it enters into force, the original text of the law is the legal baseline. And postponing the high-risk requirements is not the same as cancelling them: anyone who inventories and trains now will not have to rush later.

Rights of pupils and parents

The law strengthens the position of those who are assessed by AI. Once Article 86 applies to the relevant high-risk AI system, pupils, students and, where applicable, their parents or legal representatives can ask for a clear and meaningful explanation of the role of that system in a decision that affects them. On top of that, the GDPR already applies: for solely automated decision-making, including profiling, with legal effects or similarly significant consequences, strict conditions and rights apply, such as human intervention, the right to express a point of view and the right to contest the decision. For schools, that means: objection procedures and school regulations must be ready for parents to start asking questions about the role of AI in an assessment or admission decision. Anyone who is transparent about this now avoids awkward conversations later.

Supervision

Formal AI Act supervision will run through designated national supervisory authorities. In addition, questions about AI use may also come up in existing supervisory and accountability relationships, for example around privacy, educational quality, examination policy or complaints handling. Demonstrability is key here: showing which systems you use, how human oversight is arranged and how staff have been trained.

The tension that educational institutions feel

What makes educational institutions unique is that AI does two things at once. On the one hand, adaptive learning, personalised feedback and automated practice can have real didactic value. On the other hand, using AI at school touches directly on equal opportunity and the life paths of pupils and students. An algorithm that systematically rates pupils from disadvantaged neighbourhoods lower, or that classifies pupils with autism as "deviant behaviour" during an online exam, that is not a technical problem, that is a justice problem.

The EU AI Act is written above all with that in mind. Not to stop schools from using AI but to make sure that use stays responsible, with human oversight, with transparency towards pupils and parents and with demonstrable checks on bias.

The four roles that each need their own training

One generic "AI training for the school team" is not enough to take Article 4 seriously, certainly not in a context where AI helps decide about pupils. In practice, four groups can be distinguished:

A school that does not address these four roles separately will find it hard, when the inspectorate or parents ask, to demonstrate that Article 4 has been taken seriously.

The forgotten group: the pupils themselves

Strictly speaking, Article 4 is aimed at the institution's staff. But pupils and students use AI en masse themselves, for homework, for assignments, for dissertations. Schools that ignore this miss half the story. It is wise to put AI literacy on the curriculum for pupils as well, not as an AI Act obligation but as part of citizenship and digital skills in a world where AI is not going away.

What you can realistically do now

In closing

Education is the place where young people learn to think about the world they will soon work and live in. If anyone needs to understand well how AI works, where the limits lie and how to use it responsibly, it is the next generation. Educational institutions that invest in AI literacy now therefore do much more than compliance, they give their pupils and students a skill that will matter for the next twenty years.

At AIAdopt, we developed the Education sector extension (M3-ON) specifically for teachers, school leaders and support staff in primary, secondary and higher education. The training covers bias in assessment systems, human oversight in admission and examinations, the dialogue with pupils and parents about AI use and the interplay between the AI Act and the GDPR, ending with an assessed certificate that lists the AI Act articles covered. A certificate is not a legal obligation but a practical piece of evidence that helps demonstrate which knowledge has been covered and tested.

👉 See our approach for education or build your own package tailored to your institution.