An AI policy for your school: the decisions school leaders cannot skip

Why an AI policy now?

Pupils already use AI, but school policies are often missing or unclear.

Stanford HAI’s 2026 AI Index Report puts the gap in numbers. Over 80% of US high school and college students now use AI for school-related tasks. Yet only half of middle and high schools have AI policies in place, and just 6% of teachers say those policies are clear. Stanford sums it up as: formal education is lagging behind AI. The gap is not new. In 2023, a UNESCO global survey of over 450 schools and universities found that less than 10% had institutional policies or formal guidance on generative AI, largely because national rules were missing.

In Europe, the rules are no longer missing. Parts of the EU AI Act already apply to schools, and more will follow (see below). A policy is how a school turns those rules, and its own values, into everyday choices for teachers and pupils. For inspiration on what schools are already doing well, see how schools are using AI to transform learning and the role of generative AI in modern education.

Let children and students learn to work with AI, because it is never going away. You can ban it, but AI is here to stay, so guiding them works better than banning it.

Decision 1: What may AI do?

First agree who decides, then agree per task what AI may do and what stays with the teacher.

Start with people, not tools. Agree who in the leadership team decides what is allowed, who keeps watch and how teachers learn to work with AI. That is the core of an AI-first organisation, in a school too. Next, look at the work itself. Go through the tasks with your staff, from lesson preparation and marking to timetabling and preparing parent meetings, and agree per task: what may AI do, and what stays with the teacher? Then make clear what is allowed and for what. Teachers need to know what they can use for lesson preparation. Pupils need to know what they can use for homework, and when they cannot. Age matters. UNESCO’s Guidance for generative AI in education and research (2023) sets an age limit of 13 for the use of AI tools in the classroom, and proposes an age-appropriate approach. UNESCO also warns that many educational institutions are “largely unprepared to validate the tools”. So agree who checks a new tool before it is used with pupils: who made it, what data does it collect, and what happens to that data?

Some uses are simply off limits. Under Article 5 of the AI Act, AI systems that infer the emotions of people in education institutions are prohibited, except for medical or safety reasons. According to the European Commission, this ban has applied since February 2025. A tool that claims to measure how engaged or stressed pupils are by reading their faces does not belong in a European classroom.

Decision 2: How to assess now?

If you only see the finished essay, AI blurs what a pupil can do alone, so choose an approach per type of task.

This is the decision most schools put off, and the one teachers ask about first. AI can write a take-home essay in seconds. If the school only sees the finished product, and not how it was made, it is hard to tell what a pupil can do on their own. Schools have three broad options, and most combine them: assess more of the process (drafts, oral explanations, work in class), allow AI explicitly and ask pupils to show how they used it, or keep certain tasks AI-free and supervised.

Research shows why ‘without AI’ belongs in the mix. According to a Stanford review from 2026, only 20 of more than 800 studies on AI in education provide strong causal evidence. The pattern: pupils perform better while they use AI, but without AI the effects are mixed. In an experiment in Turkey, secondary school pupils who practised maths with a general chatbot did worse on a closed-book exam than pupils without AI; with a chatbot that acted as a tutor and gave hints instead of answers, they scored the same. So also assess without AI what a pupil must be able to do alone. Whatever you choose, write it down per type of task, so pupils do not get different rules from every teacher. We looked at the student side of this in almost every college student is cheating with AI, so now what?

There is also a legal line. The AI Act classifies certain AI systems in education as high-risk. According to Annex III, these include AI used to decide admission, to evaluate learning outcomes, to assess the level of education a pupil will receive, and to monitor pupils for prohibited behaviour during tests. The European Commission says strict obligations for these high-risk uses apply from 2 December 2027, including human oversight. A school that lets AI grade exams, or uses AI proctoring, should know this before signing a contract.

Decision 3: Pupil data

Agree which data may go into which tools, who checks a tool first and who to contact when something goes wrong.

What users type into an AI tool can end up with the provider. The Dutch education technology foundation Kennisnet advises schools to make sure the provider may not use prompts or uploaded files as training data. In a school, that can mean names, grades, special needs or family situations. UNESCO’s guidance warns that without proper rules, the data privacy of users is left unprotected, and calls for data protection and privacy standards. A policy can make three things concrete:

• which data staff and pupils may enter into which tools (for example: no personal data about pupils in tools the school has not approved);
• who checks a tool with the school’s data protection officer before it is used;
• who to contact when something goes wrong.

Decision 4: AI literacy

AI literacy has been an EU obligation since 2 February 2025, and schools do not have to start from scratch.

A policy only works if people understand the tools. In the EU, this is also a legal point. Article 4 of the AI Act asks organisations that deploy AI systems to take measures to support the AI literacy of their staff. According to the European Commission, this has applied since 2 February 2025, and national market surveillance authorities supervise it from 2 August 2026. The rule was softened this year. Since the Digital Omnibus entered into force on 27 July 2026, AI literacy remains an obligation, but no specific or “sufficient” level is required. There is no need for a certificate either: organisations can keep an internal record of trainings. For schools that use high-risk AI systems, the obligation to train staff for human oversight remains in place.

For the content, schools do not have to start from scratch. On 18 June 2026, the European Commission and the OECD presented an AI Literacy Framework for primary and secondary education. It is meant for teachers, education leaders and policymakers and is organised around four domains: engage with AI, create with AI, manage AI and shape AI. The Commission’s Q&A notes that the framework is non-binding. UNESCO also calls for teacher training. In practice, a short session per team on what AI does well, where it fails and how people keep watch is a better start than a long course nobody finishes.

Decision 5: Who updates it?

Name an owner and review the policy at least once a school year.

AI tools change every few months. A policy written once and filed away is out of date within a year. Give that task to one person, for example a member of the leadership team, with a fixed date in the diary. Involve teachers, pupils and parents: they see in practice where the policy works and where it does not. Mistakes are part of this; our article on the six biggest mistakes schools make when embracing AI shows the most common ones.

A one-page AI policy

A workable policy fits on one page, in six parts.

Put your decisions under these six headings:

1. Purpose.
   Why the school uses AI, and what it wants to protect (learning, fairness, privacy).
2. Agreements on use.
   Who decides what is allowed, per group (staff, pupils by age), and for what.
3. Red lines.
   No personal data in unapproved tools; no emotion recognition; no AI as the only judge of a grade.
4. Assessment.
   Per type of task: AI allowed, allowed with explanation, or not allowed.
5. Training.
   What every staff member learns, and how it is recorded.
6. Owner and review.
   Who is responsible, and when the policy is next reviewed.

Frequently asked questions

Should schools ban AI?

A ban is hard to enforce when over 80% of US high school and college students already use AI for schoolwork, according to Stanford HAI. UNESCO argues for a human-centred approach with clear rules, age limits and teacher training, rather than ignoring the tools. Some uses, like emotion recognition, are banned in the EU anyway.

Is AI in schools high-risk?

Not all of it. Using AI to prepare lessons is not on the high-risk list. AI that decides admission, evaluates learning outcomes, sets the level of education or monitors pupils for prohibited behaviour during tests is in principle high-risk under Annex III. There is an exception in Article 6(3): a system that only performs a narrow procedural or preparatory task, without materially influencing the outcome of a decision, is not high-risk. A system that profiles pupils is always high-risk. Strict obligations for those uses apply from 2 December 2027.

Do teachers need an AI certificate?

No. The European Commission states there is no need for a certificate. Schools that use AI must support the AI literacy of their staff, and can keep an internal record of trainings.

From what age can pupils use AI tools?

UNESCO’s guidance sets an age limit of 13 for the use of AI tools in the classroom. That is advice from UNESCO, not a general legal minimum age. Check the age rules of each tool as well, and set your own limits per age group in the policy.

Where do school leaders start?

Put the five decisions on the agenda of one leadership meeting and make a first choice for each.

Plan one leadership meeting for it. Make a first choice for each decision, write it on one page, and share it with the team as a draft. Ask teachers and pupils for feedback within a month. A clear, imperfect policy helps more than a perfect one that never arrives.

A one-page policy is a start, but whether it works depends on the people in the school. I recommend that school leaders give teachers room to try things and experiment with new technology, within the school’s agreements, and then discuss together what works in the classroom and what does not. Teachers who are enthusiastic and critical at the same time pass that on to their pupils. How to get people moving in that way is also the subject of the keynote on the organisation of the future. How AI is changing learning, work and society is also a theme in the keynote on the AI revolution.