The short answer
Skolverket has published guidance on AI, chatbots and similar tools. It does not forbid them. It sets conditions, and the conditions are more specific than most of the debate around them.
Four of them matter most. The use has to comply with the EU AI Act. Personal data and student texts should not be handed over. The teacher’s professional judgement, and any decision about grades, cannot be replaced. And schools should have written guidelines they can revise as they learn.
None of that rules out using AI to help students who need support. It does decide what a serious tool has to look like.
What the guidance actually says
On the school’s own responsibility: “En huvudman eller en skolenhet som vill använda AI-baserade lärverktyg behöver se till att användningen uppfyller EU:s AI-förordning.” The duty sits with the huvudman or the school unit, not only with the supplier. Buying a tool does not outsource the obligation.
On data, the sentence worth reading twice: “Lämna därför inte ut personuppgifter eller elevtexter utan tillåtelse. Det går inte att radera information som en AI har fått tillgång till.” The second half is the part that catches people. Information given to an AI service cannot reliably be taken back, which makes this a decision to get right the first time rather than one to correct later.
On the teacher’s role: “AI kan användas för att stödja undervisning och lärande, men får inte ersätta lärarens professionella kompetens eller beslut om betyg.” Support, not substitution. Skolverket also advises against using submitted assignments as a basis for grading without checking them.
On critical use: “AI-genererade texter kan vara övertygande men ändå innehålla fel.” Convincing and wrong at the same time is exactly the failure mode a school has to plan for, which is why students need to be able to examine AI-generated content rather than receive it.
The part schools have not priced in yet
Skolverket’s first condition points at a law, and that law has a calendar.
The AI Act entered into force on 1 August 2024. The prohibitions and the AI literacy obligation have applied since 2 February 2025, so the duty to make sure staff have adequate AI competence is not something in the future. General-purpose AI obligations followed on 2 August 2025.
The obligations for high-risk systems in areas including education were revised through the AI Omnibus amendments and now apply from 2 December 2027. That is a delay from the original date, and the practical reading is not “we can stop thinking about it”. It is that schools and suppliers have a defined window to get the documentation, oversight and data governance in order before the rules bite.
What “high-risk” covers in education
The category is narrower than “any AI in a school”, and knowing where the lines fall is what lets a school ask a useful question instead of a vague one.
Under the AI Act’s Annex III, education uses treated as high-risk include systems used to determine access or admission to education, to evaluate learning outcomes, to steer the learning process, to assess the appropriate level of education a person should receive, and to monitor and detect prohibited behaviour during tests.
Read that list against the tools already in a Swedish school and it becomes a procurement question rather than a philosophical one. A tool that grades, sorts, or decides what a student is capable of sits closer to those categories than one that helps a teacher prepare and document support.
What this means for a support tool
For students who need extra anpassningar or särskilt stöd, the useful division of labour follows directly from Skolverket’s own wording.
The decision stays human. Whether a student needs särskilt stöd, what goes in the åtgärdsprogram, and what a student’s work is worth are judgements the law places with people, and Skolverket says plainly that AI may not replace them.
What can be delegated is the preparation and the record. Adapting material so a student can reach the same goals, and carrying the documentation burden that otherwise eats the time meant for teaching, are exactly the tasks where a tool earns its place without touching the decisions it should not be making.
That distinction is not a marketing position. It is the shape Skolverket’s guidance leaves available.
Questions worth asking any vendor
If you are evaluating an AI tool for your school, these are the questions the guidance implies. They are worth asking us too.
- What personal data does the tool process, where is it stored, and who is the personuppgiftsbiträde?
- Can data be deleted, and what exactly is retained after deletion?
- Does the system evaluate learning outcomes, steer the learning process, or influence access decisions, and if so, how is it positioned against the AI Act’s high-risk categories?
- Where does a human make the decision, and is that documented?
- What does the tool do when it is uncertain or wrong, and how would a teacher notice?
- What does the supplier provide to help the huvudman meet its own obligations?
A supplier who cannot answer these is asking the school to carry a duty it cannot discharge.
Where Nuro fits
Nuro is built for the side of the line Skolverket leaves open. It adapts material to how a student learns and turns what happens into documentation the school can actually use, while the teacher keeps the professional judgement and the decisions.
We think schools should ask us the questions above rather than take that description on trust. The wider policy debate about AI in Swedish schools has mostly been about students using AI to write. The question underneath it, whether AI should be used to deliver the support the law already requires, is the one we are trying to answer, and it is a question special education has its own perspective on.