The education sector is under pressure. In Sweden, hundreds of students have been expelled for AI cheating. In Denmark, one in five examiners report problems with AI in assignments, and several universities have scrapped take-home exams in favor of pen and paper.
Here is the first newsletter of the week, which addresses the technical and structural frameworks for AI. On Wednesday we will send the final newsletter of the year, which looks more closely at news with a more philosophical angle on how AI affects the relationship between people.
This newsletter instead focuses on the system, the new tools and rules, and the challenges that compel us to change practice.
A positive piece of AI news in Denmark is that Ordbogen has just launched Chat.dk, a Danish language model built on open source and hosted in Denmark. We have tested it ourselves and are quite impressed with how good it is in Danish.
Diving into the other news, however, a picture emerges of a pressured education sector. This is not only the case in Denmark. In Sweden, hundreds of students have been suspended, because they cheated using AI. Here in Denmark, one in five history examiners experience issues with AI in assignments. At the same time, Brian Juul is calling for clear national guidelines for the use of AI in upper secondary schools (gymnasiums).
At the universities, we see the consequence clearly. In several places take-home assignments are now being dropped in favour of in-person exams with pen and paper. In other words, we are beginning to rethink the education system in order to handle the technology.
At the same time, postdoc Anton Danholt Lautrup warns about the risk of synthetic data, where AI is trained on its own output and gradually loses its connection to reality.
Meanwhile, there is news from the media sector, where DPCMO has concluded a new agreement that ensures both control and compensation for the use of content.
But how do we move beyond bans and expulsions? In this week’s commentary I argue that we need to reset the system.
Happy reading.
Ordbogen A/S has launched chat.dk, a Danish-developed AI assistant built on Odin-LLM. The model is an open-source language model that has been trained and fine-tuned specifically for Danish. This gives it an edge for the Danish language and Danish contexts, where the large international models often have weaknesses.
The solution is run from a Danish data centre with a focus on secure data storage and compliance with Danish legislation.
Chat.dk is free for educational use and is also available as an app. Companies, municipalities and other public institutions can purchase access to direct API calls if they wish to use the assistant professionally.
Joachim Næshave
A Ritzau news item, shared across several newspapers, reports that since 2023, 770 students have been suspended from Swedish universities for using AI to cheat.
In 2025 alone, this amounts to 467 suspensions, according to a new tally from the Swedish membership magazine Universitetsläraren. In addition, there have been 75 warnings for use of AI. Stockholm University tops the list with the most cases, which is attributed both to the university’s size and to increased attention to the problem. Lecturers there have received courses, workshops and guidance on the correct use of AI.
Other universities, including Linköping, also report more AI-related disciplinary cases and adjustments to teaching and examination formats. In total, 27 of the 35 higher education institutions surveyed have changed their examination formats as a result of AI.
Sweden is thus one of the countries in Europe where universities are reacting most systematically to AI-related cheating.
Ritzau
Linus Hellerstedt
Folkeskolen.dk writes that one in five history examiners on the teacher education programme this year have recorded the use of AI in exam assignments. Usage has risen markedly since last year, and in half of the cases it has caused problems in assessment.
Rune Christiansen, chair of the corps of examiners in history, points out that there is a lack of common guidelines for how AI may be used in assignments.
History is particularly vulnerable to AI-enabled cheating because source criticism is a core competence. Many of the problematic cases are due to missing citations or so-called source hallucinations, where AI invents articles that do not exist. Examiners also spot suspicious use of AI when the language changes character or when the assignment appears incoherent.
Andreas Brøns Riise
TV 2 has spoken with principal Brian Juul about the post he shared on LinkedIn, where he expresses concern about students’ use of AI in teaching. At Slotshaven Gymnasium, 11 third-year students have received written warnings for using AI as a shortcut in connection with a major written assignment. Brian Juul assesses that the assignments are written in a language and with concepts that the students themselves would not use. He sees this as a sign that AI is being used without the students understanding the content.
Brian Juul believes the problem is not limited to his school. He finds that many gymnasiums face the same challenge, and he points out that there is a lack of clear political guidelines for how AI should be used in teaching. This creates uncertainty and an inconsistent approach across schools.
The Minister for Children and Education, Mattias Tesfaye, acknowledges the issue. He says that various initiatives have already been launched. These include exams where students must use paper and pencil. In addition, new measures are being tested locally before it is decided whether they should be introduced nationwide.
Read Brian Juul’s full call to action on TV 2:
Ulrikke von der Recke Beckmann
Djøfbladet reports that several programmes at Aarhus Universitet (Aarhus University) are changing their examination formats to address the challenges posed by AI. Take-home assignments are increasingly being replaced by in-person written exams and oral examinations. This is both to prevent cheating and to protect students from unfounded accusations of AI use.
The university has abandoned AI-detection tools because they are not accurate enough and is instead relying on invigilation and control in in-person exam situations.
Lecturers and students are calling for clearer guidelines on when and how AI may be used in assignments and examinations. At the same time, AI is becoming a natural part of teaching going forward, and the university is therefore working to integrate the use of AI into course descriptions and programme regulations.
Julius Krause
Anton Danholt Lautrup, a postdoc at the Institut for Matematik og Datalogi (Department of Mathematics and Computer Science) at SDU (the University of Southern Denmark), has just completed his PhD thesis on synthetic data.
Here he warns that large volumes of AI-generated content can lead to a form of digital inbreeding, where AI models are trained on data that no longer originates from real people. This can result in what he calls model collapse, where AI systems lose the ability to generate useful, trustworthy and diverse outputs.
He calls for increased awareness of the use of AI-generated data, especially when used in searches, automatic text suggestions and digital posts that can resemble facts without actually being so.
For Anton Danholt Lautrup, the question is no longer what AI can do, but what we choose to use it for and what consequences this has for the information society.
Read more at SDU:
birssdu.dk
Lautrup, Anton Danholt
DPCMO, Danske Pressepublikationers Kollektive Forvaltningsorganisation (Danish Collective Management Organisation for Press Publications), together with the US company ProRataAI, has entered into a collective licensing agreement for generative AI. Danish media thereby gain control over how their content is used in AI systems, as well as assurances of proper attribution and financial compensation.
As part of the agreement, the media will have access to Gist Answers, ProRataAI’s answer engine based on RAG technology. The solution can be integrated directly into the media outlets’ own platforms and is made available free of charge and on a voluntary basis to all DPCMO members.
By: Karen RøndeIf you have been following our newsletters at Viden.AI, you could easily get the impression that the education system is on the verge of collapse. That we have lost control over assignments written with AI, and that pen and paper have become the only solution.
For years we have digitised teaching and examinations on the basic assumption that access to the internet and digital aids would lead to better assignments. That assumption no longer holds. When a chatbot can solve an assignment to a top grade (12 with a star) in a few seconds, we are no longer measuring the student’s abilities, but the machine’s.
But pen and paper are not necessarily the solution. In an opinion piece in Gymnasieskolen RUC (Roskilde University) rector Hanne Leth Andersen, together with the chairs of the German and French teachers’ associations, argue that we should reconsider the very aim of foreign language teaching. They write, among other things:
"It may sound controversial, but being able to write on one’s own, especially with paper and pen, is less relevant. Quite simply, it is not a core foreign-language competence in modern society."
Instead, they point to oracy, reading and source criticism as key competences. They believe that traditional written examinations kill motivation, and that technology only makes the problem more apparent.
Having students write by hand is an emergency brake, but not a long-term strategy. It is merely a pause that gives us time to think.
The solution is therefore not to hunt down cheats or to ban the technology. Instead, we must reinvent the way we do school. We need to move away from tasks that AI can solve on its own and instead create a practice in which the process is more important than the product.
