The EU changed the timetable for the AI Act six days before it was due to apply in full. What does that mean for you as a school organiser or principal?
For school organisers, education administrations, principals and school leadership.
For almost two years, 2 August 2026 has been the date around which all planning for the EU AI Act was organised. It was the day the requirements for high-risk AI systems were due to start applying – and education is explicitly on that list (European Parliament and Council 2024, Annex III, point 3).
Six days before that, the EU changed the timetable.
That does not mean Sunday 2 August is unimportant. It means that part of the rules starts applying then, and that the heaviest part has been given another sixteen months. The difference is worth understanding precisely, because it determines what you need to have in place this week and what belongs to the 2027 budget year.
The short version
- Regulation (EU) 2026/1744 entered into force on 27 July 2026 and amends the AI Act. It is known as the Digital Omnibus on AI.
- The high-risk requirements for AI in education have been postponed from 2 August 2026 to 2 December 2027.
- The transparency requirements in Article 50 have not been postponed. They start applying on 2 August 2026.
- Two things have already applied since February 2025: the prohibition on emotion recognition in educational settings and the requirement for AI literacy among staff.
- The GDPR applies in parallel and has not moved. The school organiser is the data controller.
What happened in July
In November 2025, the Commission proposed an amendment to the AI Act, after it had become clear that the technical standards and supporting documents on which the high-risk requirements rest would not be ready in time. After a failed round of negotiations in April, a political agreement was reached on 7 May, Parliament voted in favour on 16 June and the Council gave its final approval on 29 June.
The act is dated 8 July, was published in the Official Journal of the European Union on 24 July and entered into force on the third day following publication – 27 July 2026. The legislator justifies the unusually short time to entry into force on the grounds that 2 August was so close that a longer wait would have created legal uncertainty (European Parliament and Council 2026).
So from 27 July, the AI Act is no longer the 2024 text. It is the 2024 text as amended.
The new timeline
| Date | What applies |
|---|---|
| 2 February 2025 | Prohibited AI practices (Article 5) and AI literacy requirements (Article 4) |
| 2 August 2025 | Rules for general-purpose AI models, governance and penalties |
| 2 August 2026 | The transparency requirements in Article 50 and general application of the Regulation |
| 2 December 2026 | The labelling requirement in Article 50(2) for systems already on the market; new prohibitions |
| 2 December 2027 | High-risk requirements for stand-alone systems in Annex III – including education |
| 2 August 2028 | High-risk requirements for AI embedded in products under Annex I |
Source: European Parliament and Council (2024, Art. 113); European Parliament and Council (2026).
Why schools are singled out
Annex III to the AI Act lists eight areas in which AI systems are classified as high-risk. Point 3 concerns education and vocational training and covers systems used to
- determine access or admission to education,
- evaluate learning outcomes – including when the outcomes are used to steer the student's further learning,
- assess the level of education a person will receive or will be able to access, and
- monitor and detect prohibited behaviour during tests.
(European Parliament and Council 2024, Annex III, point 3.)
This is where the most important question of interpretation for Swedish schools lies, and it has not been settled.
The wording of the second bullet deserves a careful reading. It extends the scope for systems that evaluate learning outcomes, so that they are covered even when the evaluation is used to steer the student's further learning. It does not, however, turn every tool that supports formative work into an evaluation system. The crux is whether it is the system or the teacher that carries out the evaluation.
The distinction is not academic. In its position paper, Sveriges Skolledare (the Swedish union for school leaders) points out that schools use many AI systems that are classified as high-risk under the AI Act, and mentions systems that give grade recommendations, steer admissions or disciplinary decisions, or analyse student data to provide support (Sveriges Skolledare 2025). The examples have one thing in common: the system sits close to a decision about the student.
What is decisive, then, is not what the tool is called or how it is marketed. What is decisive is which function the system actually performs, what it produces, who makes the decision – and what consequences the outcome has for an individual student.
The exception in Article 6(3) – and why you should not rely on it
Article 6(3) contains an exception. A system referred to in Annex III shall not be considered high-risk if it does not pose a significant risk to health, safety or fundamental rights, including by not materially influencing the outcome of decision-making. The exception may apply if the system
- performs a narrow procedural task,
- improves the result of a previously completed human activity,
- detects decision-making patterns or deviations without replacing or influencing a previously completed human assessment without proper review, or
- performs a preparatory task to an assessment.
The exception never applies if the system performs profiling of natural persons (European Parliament and Council 2024, Art. 6(3)).
On 19 May 2026, the Commission published draft guidelines on high-risk classification. Its position there is that AI in education often involves profiling and that the exception will therefore in many cases not apply. Narrow preparatory tasks can still qualify – systems that evaluate a student's progress in knowledge over time generally do not (European Commission 2026a).
This means that AI tools in schools end up in three different positions, which should not be treated as one.
Systems that decide or recommend. Grade recommendations, admissions support, automated marking that delivers a judgement. These perform the functions in point 3 directly. In practice, the exception is not relevant.
Systems that model the student over time. Adaptive platforms that build a persistent profile of the student and adapt content and level accordingly. This is where the Commission's reasoning on profiling bites hardest, and it is also where the industry has tried to have a narrower reading established: in 2025, the US industry association SIIA asked the Commission to clarify that points 3(b) and 3(c) refer to summative assessment and not adaptive lessons (SIIA 2025). No such clarification has been given.
Systems that draft material for a human assessment. Here, conditions (b) and (d) in Article 6(3) are the relevant ones – improving the result of a previously completed human activity, or performing a preparatory task. The assessment stands or falls on whether human oversight is real: can the teacher depart from the suggestion without friction, and is it the teacher who actually makes the judgement? If the answer to both is yes, the circumstances are different from those in the two previous positions. If the system at the same time builds a persistent profile of the student, the exception is closed regardless.
The conclusion for those of you who procure: do not accept "we are formative, so we are exempt" as an answer. The exception is an assessment that must be made per function, documented, dated and revisited when the product changes. A supplier that has done that work can show it to you. A supplier that has not, cannot.
Three things that already apply to you as a school organiser
Emotion recognition is prohibited, not high-risk. Since 2 February 2025, Article 5(1)(f) has prohibited the use of AI systems to infer the emotions of natural persons in educational settings, except for medical or safety reasons (European Parliament and Council 2024, Art. 5(1)(f)). The Commission's guidelines explicitly cite an educational institution using emotion recognition to assess students' interest and attention as a prohibited example (European Commission 2025). The prohibition targets systems that infer emotions from biometric data; systems that interpret sentiment in written text fall outside this particular provision, but are still covered by the GDPR. The maximum fine for prohibited practices is the highest in the entire Regulation.
The AI literacy requirement has applied since February 2025. Article 4 is addressed to both providers and deployers – and a school organiser that uses an AI system is a deployer. The requirement covers staff and other persons dealing with the systems on your behalf, and is to be adapted to their prior knowledge and to the context. The July amendments shifted the wording towards supporting the development of AI literacy rather than ensuring a particular level (European Parliament and Council 2026). The requirement remains.
It is worth setting this against how AI is actually being used. In the latest status report from Skolverket (the Swedish National Agency for Education), almost eight in ten teachers surveyed had used AI services in some part of their work, and seven in ten had used them to plan and prepare teaching. Four in ten said they had initiated or approved student use of AI, compared with just under two in ten in the corresponding 2024 survey. The report is based on responses from 368 teachers in Skolverket's online panel in compulsory school, preschool class and school-age educare, and Skolverket itself points out that it does not give a nationally representative picture (Skolverket 2026). On the leadership side, only four in ten school leaders say they have sufficient training to use AI safely (Sveriges Skolledare 2025).
The GDPR has not moved. The data controller is the school organiser – not the principal and not the individual teacher. In municipal schools, it is the responsible committee (Integritetsskyddsmyndigheten n.d.). That means a legal basis, data processing agreements, records of processing and, in many cases, a data protection impact assessment before an AI tool is used on student texts. IMY (the Swedish Authority for Privacy Protection) has named children and AI as priority areas for 2026 (Integritetsskyddsmyndigheten 2026a).
From 2 August: Article 50
The transparency requirements are not being postponed. From 2 August 2026, anyone interacting with an AI system must be informed of that, and providers of systems that generate synthetic content – including text – must ensure that the output is marked in a machine-readable format and detectable as AI-generated (European Parliament and Council 2024, Art. 50).
There is a limited transitional period: systems that were on the market before 2 August 2026 do not need to meet the labelling requirement in Article 50(2) until 2 December 2026. Content generated before 2 August does not need to be labelled retroactively (European Commission 2026c). In June 2026, the Commission published a voluntary code of practice on the transparency of AI-generated content that can be used to demonstrate compliance (European Commission 2026b).
In practical terms: ask your existing suppliers what changes on their side on Sunday, and ask for the answer in writing.
Who supervises in Sweden
The Swedish government inquiry on the AI Act submitted its report Anpassningar till AI-förordningen (Adaptations to the AI Act) on 6 October 2025. The model set out there means that eleven authorities share market surveillance, with Post- och telestyrelsen (the Swedish Post and Telecom Authority, PTS) as the coordinating and lead authority, while IMY is responsible for areas including law enforcement, migration, the administration of justice and biometrics, and Finansinspektionen (the Swedish Financial Supervisory Authority) for banking and insurance (SOU 2025:101). AI in schools therefore falls under the general arrangement.
Pending supplementary Swedish legislation, the Swedish Government has tasked PTS, IMY, Finansinspektionen, Läkemedelsverket (the Swedish Medical Products Agency) and Swedac (the Swedish Board for Accreditation and Conformity Assessment) with acting as national competent authorities up to and including 31 December 2026 (Swedish Government 2026; Post- och telestyrelsen 2026).
Six questions to bring into procurement
- Which of the functions in Annex III, point 3 does the system perform – and who sets the grade? Ask for a functional answer, not a marketing one.
- Is there a documented and dated classification assessment under Article 6? May we read it?
- Does the system profile students? If so, the exception in Article 6(3) is closed, regardless of what else the system does.
- What does human oversight look like in practice? Can the teacher depart from the suggestion without friction, and is it possible to see afterwards that a human actually made the judgement?
- Where is personal data processed, is there a data processing agreement, and are any models trained on our students' data?
- What changes on your side on 2 August and on 2 December 2026, respectively, to comply with Article 50?
How we work at Gradr today
Gradr suggests wording for formative feedback on student work. The teacher chooses, edits or rejects each suggestion. The evaluation is done by the teacher. Gradr sets no grades, produces no basis for grading and makes no summative assessment.
Personal data is processed on servers within the EU. Customer data is not used to train foundation models. The school organiser owns its data.
We carry out our own classification assessment per function and follow the Commission's work on the guidelines. We do not treat "formative" as an automatic exception from Annex III – as set out above, that is not what the law says today – and we revisit the assessment when the product changes significantly.
The six questions above should be put to us as well.
Published 30 July 2026. This text describes the legal position as at the date of publication and does not constitute legal advice. The legal position on the classification of AI in education is evolving; at the time of publication, the Commission's guidelines on high-risk classification were a draft.
References
European Parliament and Council (2024) Regulation (EU) 2024/1689 of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union, 12.7.2024.
European Parliament and Council (2026) Regulation (EU) 2026/1744 of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplified implementation of harmonised rules on artificial intelligence. Official Journal of the European Union, 24.7.2026. ELI: reg/2026/1744.
European Commission (2025) Guidelines on prohibited artificial intelligence practices established by Regulation (EU) 2024/1689. C(2025) 884 final, 4 February 2025.
European Commission (2026a) Draft guidelines on the classification of high-risk AI systems under the AI Act, 19 May 2026.
European Commission (2026b) Code of Practice on Transparency of AI-generated Content, June 2026.
European Commission (2026c) Transparency obligations under Article 50 of the AI Act – questions and answers. Shaping Europe's digital future.
Integritetsskyddsmyndigheten (2026a) IMY:s prioriteringar 2026 – AI, barn och brottsbekämpning [IMY's priorities for 2026 – AI, children and law enforcement], 3 February 2026.
Integritetsskyddsmyndigheten (2026b) IMY blir marknadskontrollmyndighet för AI-förordningen [IMY becomes a market surveillance authority for the AI Act], 15 June 2026.
Integritetsskyddsmyndigheten (n.d.) För personuppgiftsansvariga inom skola och förskola [For data controllers in schools and preschools].
Post- och telestyrelsen (2026) AI-förordningen [The AI Act].
Swedish Government (2026) Uppdrag att vara nationella behöriga myndigheter enligt AI-förordningen [Assignment to act as national competent authorities under the AI Act]. Government decision, ref. Fi2026/01365, published 12 June 2026.
SIIA (2025) SIIA Seeks Clarification on Implementation of EU AI Act's Rules for High-Risk AI Systems in the Educational Context, 21 July 2025.
Skolverket (2026) Artificiell intelligens i undervisningen – en lägesbild över lärares användning och hantering av AI under hösten 2025 och starten på 2026 [Artificial intelligence in teaching – a snapshot of teachers' use and handling of AI in autumn 2025 and early 2026]. Stockholm: Skolverket, 26 May 2026.
SOU 2025:101 Anpassningar till AI-förordningen – Säker användning, effektiv kontroll och stöd för innovation [Adaptations to the AI Act – safe use, effective control and support for innovation]. Report of the Swedish Inquiry on the AI Act. Stockholm, 6 October 2025.
Sveriges Skolledare (2025) Ställningstagande – AI i skolväsendet [Position paper – AI in the school system], December 2025.
