Will Norway end short take-home exams because of AI? An expert panel recommends it—but no rule has changed
A national inquiry says short, unsupervised take-home exams are increasingly unreliable when generative AI is readily available. It recommends more controlled written and oral assessment, wider AI competence and strict conditions before automating grade explanations, but the proposals are advisory and now enter consultation.
The short version
- What happened
- Norway's official expert inquiry recommends reducing short, unsupervised take-home exams and combining controlled written or oral assessment with longer work where students must demonstrate independent competence.
- Who it affects
- Students could face more oral or supervised components and would need clear expectations, accessible arrangements and a fair route to challenge decisions.
- What remains unproven
- The inquiry is advisory and entering consultation; no national rule, ban or funding settlement has been enacted.
Keep in mind
The inquiry is advisory and entering consultation; no national rule, ban or funding settlement has been enacted.
It is a policy synthesis and deliberative process, not a controlled study of learning, misconduct, workload or equality outcomes.
The report does not provide a national student denominator or causal estimate for the effect of generative AI on assessment validity.
Oral and supervised assessment can create accessibility, staffing, cost and appeal challenges that require testing and mitigation.
The source is in Norwegian; this report uses an editorial translation of the official report and announcement.
Source analysis · 2 linked sources. Primary analysed source date: 6 November 2026.
Read the full report · 9 min The evidence, methods, implications and limitations
The answer: Norway has a recommendation, not a ban
Norway has not abolished take-home exams. On 6 November, a government-appointed expert committee delivered a 204-page public inquiry arguing that short, unsupervised take-home assessments have become less dependable when students can readily use generative AI outside an examiner's view. Its central recommendation is that universities and colleges should use fewer of these assessments and place more weight on controlled written or oral work that lets an institution verify what a student can do independently. The report now enters consultation; ministries, institutions and other affected groups can challenge or refine the proposals before any policy or legal change follows.
That distinction matters because the report is already consequential for lecturers designing the next assessment cycle, yet it has no direct regulatory force. Institutions retain responsibility for programmes, examinations and quality assurance under existing law. The committee proposes a direction for national policy and institutional practice, not an immediate prohibition. Calling it a ban would overstate the evidence and the government's decision. The accurate present-tense finding is narrower: Norway's expert panel thinks AI has weakened the evidential value of some unsupervised assessment formats and wants a more varied system that can still test independent competence.[1][2]
What the panel wants assessment to look like
For short pieces of assessed work, the committee recommends reducing unsupervised take-home examinations and increasing controlled written and oral assessment. For bachelor's dissertations, master's theses and other extended work, it does not propose abandoning projects completed over time. Instead, it recommends adding a controlled component—such as an oral defence or supervised element—so the student can explain choices, demonstrate subject knowledge and show ownership of the submitted work. The report also says oral elements should be reviewable, including through appropriate recording, and asks the government to examine the legal basis needed for that approach.
The logic is not that every use of AI is cheating. The panel treats AI as a normal technology that can support learning and professional work, while arguing that an examiner still needs credible evidence of the student's competence. A controlled component can make the assessment question more precise: not whether a detector thinks text looks machine-generated, but whether the student can reason, explain and apply knowledge. That avoids making unreliable AI-detection scores the foundation of a misconduct decision. It also creates practical burdens—rooms, staff time, accessibility arrangements, appeals and secure recording—that institutions would need to resource rather than simply transfer to teachers.[1]
The distinction that matters
What would change if the recommendations are adopted
The inquiry links assessment redesign to competence, rights and infrastructure rather than treating AI use as a detector problem.
Assessment
Reduce short unsupervised take-home exams
More controlled written or oral evidence
Long work
Add a controlled component to dissertations
Student demonstrates ownership and understanding
Competence
Teach general and subject-specific AI use
Students and staff remain responsible for judgement
Infrastructure
Develop shared secure computing and storage
Less dependence on consumer services for sensitive work
AI competence is presented as a learning outcome, not an optional add-on
The inquiry recommends that all students develop both general AI competence and competence tailored to their subject. It similarly calls for all teaching staff to have general knowledge of AI and the educational judgement needed to use it responsibly. The report's student principles emphasise understanding how a tool works, checking outputs, protecting personal and confidential information, acknowledging permitted use and remaining responsible for submitted work. In this model, a history student, engineer and nurse would share baseline critical skills but would also need discipline-specific practice because acceptable use, evidence and risk differ across fields.
That proposal shifts the institutional task from publishing one generic chatbot rule to designing curricula, assessment and staff development together. A university cannot fairly test responsible AI use if students have not been taught what responsible use looks like, and it cannot assume lecturers have time or confidence to redesign assessment without support. The panel therefore links competence to leadership, quality systems and working conditions. For students, the practical implication is that AI literacy may become something they are expected to demonstrate, not merely advice attached to an integrity policy. For staff, it implies paid development and access to approved tools rather than private experimentation with consumer services.[1][2]
Automated explanations of grades face a deliberately high threshold
The report considers whether AI could help produce explanations for grades, an area where automation might appear to save substantial staff time. It does not recommend immediate deployment. The committee says trials should first establish that an AI-generated explanation gives a student a comparable ability to understand the decision and judge whether to appeal. It also says a legal basis would be required. Those conditions recognise that an explanation is not decorative text: it affects a student's procedural rights, confidence in marking and decision about whether to challenge an outcome.
A useful trial would therefore need more than a fluent sample response. It would need representative assessments, clear comparison criteria, human review, records of errors, analysis across subjects and student groups, and evidence about whether recipients can identify the basis of the grade. Institutions would also need to decide who remains accountable when an explanation misstates the marker's reasoning. The inquiry does not provide such outcome evidence; it sets a governance test. Until comparable understanding and legal authority are demonstrated, automated grade explanations should be treated as an option to investigate, not a validated service or an entitlement universities can safely replace.[1]
How the inquiry worked—and what its evidence cannot prove
The government appointed the committee on 21 March 2025. According to the report's process account, it held 11 ordinary meetings through 9 October 2026, arranged five open input meetings, visited the OECD in Paris and received 73 written submissions from 64 senders by 5 October 2026. The inquiry synthesises research, legal and policy analysis, institutional experience and stakeholder evidence. That is a substantial deliberative process, and the report is a primary source for what the committee recommends. It is not, however, a national experiment comparing assessment designs across students or institutions.
There is no learner denominator from which to calculate an effect on attainment, misconduct, workload or equality, and no randomised comparator showing that oral or supervised components produce better learning. Stakeholder submissions can reveal operational problems and contested values but do not estimate prevalence on their own. The panel's recommendations should therefore be read as reasoned policy judgement under uncertainty. Confidence would increase if pilots measured validity, reliability, accessibility, staff time, student stress, appeal outcomes and differences by discipline and disability. Evidence of worse learning, disproportionate barriers or unsustainable workload would weaken the case for a specific design even if the general problem remains.[1]
Infrastructure connects classroom choices to European digital sovereignty
The committee also recommends shared infrastructure for computing and storage, ideally developed through national, Nordic or European cooperation. The aim is to give researchers, students and staff access to AI systems without forcing sensitive data, unpublished work or institutional knowledge through consumer platforms on unclear terms. It calls for a governance framework that addresses responsibility, security and access. This is not merely an IT procurement detail: the tools an institution can safely offer shape which students can participate, what research data can be processed and how dependent public universities become on a small number of external providers.
Implementation will determine whether the report changes practice. Government would need to decide which recommendations to accept, clarify funding and legal authority, and coordinate with autonomous institutions. Universities would need to test assessment changes with students and staff, provide accessible alternatives, monitor unequal effects and publish what they learn. The strongest near-term reading is therefore neither that traditional exams have won nor that AI has made independent work impossible. Norway's inquiry proposes a new bargain: teach AI use openly, verify individual competence where it matters, and build public capacity for sensitive work. Consultation and measured trials will show whether that bargain is workable.[1][2]