Christian K. Karl of the University of Duisburg-Essen has presented a model in the Hochschulforum Digitalisierung that assesses the pressure from AI for each exam individually. He sees the greatest pressure for change in unsupervised written formats such as the term paper.
Oral exams, colloquia, supervised exams and practical demonstrations remain meaningful in his assessment, because the working process and the person become visible. The article reflects the position of a single subject-didactics expert and is not a requirement of any university. What is binding is what your examination regulations say.
Which exams change the most due to AI?
With his LOCAM-AI model, Karl assigns exam elements to three types. For Type A, such as scheduling or evaluating feedback, AI supports the work and the exam logic remains. For Type B, content and assessment change: reproductive tasks lose significance, while transfer, judgment and reflection gain importance. Type C concerns the implementation, especially unsupervised written formats such as term papers, seminar papers and homework assignments. There, it can no longer be assumed that the work is the student's own effort.
Which exam remains meaningful?
Assessment by Christian K. Karl, Hochschulforum Digitalisierung, 13 November 2026
Exam formatAssessment in the article
- Oral exam, colloquiumremains meaningful
- Supervised examremains meaningful
- Term paper with process and presentationremains meaningful
- Term paper without visible processPortfolio, defense or case analysis
Expert article by a single author, not a requirement of any university
Source: Hochschulforum Digitalisierung
According to the article in the Hochschulforum Digitalisierung, the term paper does not have to disappear. It remains meaningful if process documentation, reflection and a detailed oral presentation with discussion are added. If the visible process is missing, Karl suggests a portfolio, a defense or a documented case analysis as a replacement.
Karl considers it legitimate to exclude AI if the exam structure ensures this and the focus is on independent knowledge, manual skills or safety-critical judgment. His example comes from construction engineering. Because decisions in construction are relevant to safety, costs and liability, the critical review of AI results should itself be the subject of the exam there, for example in a documented case analysis of schedule risks. What is assessed is the evaluation of the data, whether AI results are adopted or rejected, and the justification.
How far along are universities with their rules?
The KI-Monitor 2025, on which Karl relies, shows how far along the rules are. According to the press release of the Stifterverband, university didactics units at public universities reported in May and June 2025: 87 percent of universities had adapted their declarations of independent authorship, but only 43 percent had adapted their examination regulations. 15 percent had adopted an AI strategy, and another 50 percent were working on one.
In Karl's view, an obligation to declare AI use is not sufficient. He refers to Corbin, Dawson and Liu (2025), who see a false sense of security in such an obligation: meaningfulness only exists when the task itself changes. The question of whether AI was used leads to a climate of mistrust that can be resolved neither technically nor legally. Karl instead asks whether an exam still demonstrates what it is supposed to demonstrate.
What this means for you
If you are a student, it is best to record your working process: drafts, interim versions and the places where you checked or rejected AI results. This is our own assessment, as the article does not give any explicit recommendations for students. It is also best to check in your examination regulations whether they regulate AI. In summer 2025, only 43 percent of universities had adapted their regulations.
For the next exam planning, Karl offers instructors five questions:
- Should the exam demonstrate knowledge, judgment or action?
- What could AI take over in the task, and does the exam afterwards still demonstrate what is intended?
- Is AI excluded, permitted or part of the competence being assessed, and why?
- Where do the working process and the person become visible, for example in a portfolio, a colloquium or a defense?
- Do the assessment criteria list result, process and individual contribution separately?
The Hochschulforum Digitalisierung published the article on 13 November 2026.
Sources
