The AI Act in Schools: Which Uses of AI Are High Risk

What Annex III says about education

Annex III of the AI Act lists the areas in which an AI system is considered high risk. For education and vocational training there are four cases:

• Determining access or admission to institutions, or assigning people to an institution or programme.
• Evaluating learning outcomes, including when those outcomes are used to steer the learning process.
• Assessing the appropriate level of education a person will receive or be able to access.
• Monitoring and detecting prohibited behaviour of students during tests, such as anti-cheating systems using video surveillance.

The logic is simple: in these cases the system's output can affect a person's educational path and therefore their life chances, and that person is often a minor.

Practical examples: what is in and what is out

• Marking and grading tests with AI: the typical use the regulation treats as high risk.
• A platform that decides who is admitted to a course or which class a student joins: high risk.
• A system that watches students during an online exam to catch cheating: high risk.
• Creating a revision quiz, generating an image for a worksheet, getting help writing the descriptors of a marking rubric: normally not high risk, because AI prepares material and assessment stays with the teacher.
• A chatbot that answers students' questions: not high risk, but students must know they are talking to an AI (Article 50).

What is banned

Since 2 February 2025 the use of emotion recognition systems in educational institutions has been banned, except for medical or safety reasons. Software that claims to measure students' attention, boredom or stress from their facial expressions is not allowed in the classroom. The other practices banned by Article 5 also apply, such as exploiting vulnerabilities linked to age.

The exception: when an Annex III system is not high risk

Article 6(3) provides that a system listed in Annex III is not high risk if it does not materially influence decisions: for example, if it performs a narrow procedural task, improves the result of a previously completed human activity or only performs a preparatory task. The exception never applies if the system profiles people. A tool that flags spelling mistakes to the teacher is very different from a system that assigns the grade.

What teachers and schools should do

• Train staff: the AI literacy obligation already applies to anyone using AI systems, schools included.
• Keep assessment in the teacher's hands: AI can suggest, but grades and judgements remain a human decision, justified and verifiable.
• Choose suitable tools: high-risk uses require systems designed and declared compliant for that purpose, not a general-purpose assistant used outside its intended purpose.
• Protect students' data: the GDPR applies, as do internal rules on what may be entered into AI tools.
• Inform students and families when AI systems that concern them are used.

National law may add further rules, for example on minors' access to AI tools: check the rules in your country.

The deadlines

The bans have applied since 2 February 2025 and the transparency obligations since 2 August 2026. The obligations for Annex III high-risk systems, education included, apply from 2 December 2027, following the postponement agreed in the Digital Omnibus. Those using these systems will, among other things, have to follow the provider's instructions, assign oversight to competent people, keep the logs and inform the people affected. Public bodies, such as state schools, will also have to carry out a fundamental rights impact assessment.