In admissions, AI can answer, gather and organise. When it scores a family, EU law treats it as high-risk, and Swiss law already asks for explanations.

Picture this: November, the admissions office of a boarding school in the Engadine. Hundreds of applications are arriving for next year, from a dozen countries. A vendor gives a demo: each file appears with a fit score from 0 to 100 and a colour. Green, amber, red. The team, tired, feels relief. Nobody in the room asks what the machine looked at to produce that number, or what the school will say to the family who wants to know why their application ended up amber.

In admissions, AI can do a lot of useful work. There is one part, however, where it stops being an office tool and becomes, under EU law, a high-risk system. The line is not drawn by the technology. It is drawn by what the tool does with the person.

What the EU regulation says.

The EU Artificial Intelligence Act, Regulation (EU) 2024/1689, lists as high-risk in its Annex III the AI systems intended to determine access or admission to educational institutions at all levels, or to assign people to those institutions. The same list includes systems that evaluate learning outcomes, those that assess the appropriate level of education for a person, and those that monitor and detect prohibited behaviour by students during tests. Following the Digital Omnibus on AI, Regulation (EU) 2026/1744, published on 24 July 2026, these obligations apply from 2 December 2027.

For anyone using one of these systems, the regulation requires, among other things, that human oversight is assigned to people with the necessary competence, training and authority, that the logs the system generates are kept for at least six months, and that people are told when a high-risk system is used to make decisions about them or to help make them.

The blind spot: scoring already counts.

The regulation provides for exceptions. An Annex III system may not be considered high-risk if it only performs a narrow procedural task, improves the result of a previously completed human activity, detects patterns without replacing human assessment, or performs a preparatory task. But it adds a sentence that changes almost everything: an Annex III system is always considered high-risk where it performs profiling of natural persons.

A fit score, a prediction of academic success or a priority ranking of applicants evaluates personal aspects of a child or a family. If that is profiling within the meaning of the regulation, and I would treat it that way until legal counsel says otherwise, then the tool that “only prepares” the committee’s decision is on the list.

What already applies in Switzerland today.

Switzerland does not yet have a specific AI law. In February 2025 the Federal Council commissioned a consultation draft by the end of 2026, focused on transparency, data protection, non-discrimination and oversight. But the Federal Act on Data Protection, in force since 1 September 2023, already has an article for this. Under Article 21, when a decision is based exclusively on automated processing and has a legal consequence or a considerable adverse effect for the person, the controller must inform them. On request, the person must be able to express their point of view and ask for the decision to be reviewed by a natural person.

The law makes an exception where the decision is directly connected with a contract and the person’s request is granted, and where the person has explicitly consented to the automation. Put another way: an admission granted by a machine does not trigger that duty. An automated rejection does. And anyone who asks the school for access to their data has the right to know whether an automated decision was taken, and the logic behind it.

What I do not know.

I do not know whether the EU regulation reaches a Swiss boarding school deciding on a family resident in the EU. It also applies to providers and deployers outside the Union where the output of the system is used in the Union, and how that fits an admission decided in Switzerland is a question for your legal counsel. Nor do I know what each vendor’s tool does internally: only its documentation can tell you that, in writing. I am not a lawyer, and this note does not replace that advice.

Sorting paperwork is not the same as sorting families.

Where I stand.

I build AI automations, and in admissions I would put them to work without hesitation on what currently eats the team’s time: answering families at any hour and in their language, gathering and checking documents, sending reminders about deadlines, scheduling interviews and summarising the file for the person who will read it in full. I would not have them score, rank or recommend a yes or a no. Not out of fear of the rules, but because a family paying what a Swiss boarding school costs expects someone with a name to have read their child’s application. AI can prepare the decision. Signing it is a person’s job.

What I would check.

• Every tool in the process. Form, CRM, assistant, testing platform: whether any of them produces a score, a prediction or a ranking.
• Online placement tests. Whether they monitor the test automatically or assign a level: both are on the Annex III list.
• Rejections. Whether any goes out without a person having decided it, for example an automatic email for not meeting a criterion.
• The contract with the vendor. What the system does, which logs it keeps, where and for how long.
• The admissions page. Whether it says who decides, what AI does, and how to ask for a person to review a case.

Where to start.

1. This week

   Take an inventory

   A list of every tool that touches an application, with what it does in one sentence.

2. This month

   Ask in writing

   Ask each vendor whether its system scores, ranks or monitors tests, and with what data.

3. This quarter

   Say it on the website

   One clear sentence on the admissions page about who decides and how to request a review, in every language.

4. Before December 2027

   Document the oversight

   Who reviews, with what training, and what is logged, if any tool turns out to be high-risk.

Put yourself in the place of a family who receives a no. Polite, quick, with no explanation. Now ask yourself a question perhaps nobody at your school has asked yet: at what point in that process did a machine take part, and who could explain it to that family if they asked?

If the answer is “at none”, check it with your provider. If it is “I don’t know”, this is the note your admissions team should read this week.

If you would like me to review what AI does in your admissions process and how your website describes it, start with the Strategic Diagnostic. Whether a tool is high-risk is for your legal advisers to decide, not a designer.