Why Exam Regulations Must Be Updated
Most exam regulations at Swiss universities date from a time when "aids" meant handwritten notes or at most a calculator. The core phrase — "use of unauthorized aids constitutes attempted cheating" — works for proctored exams. For seminar papers written on personal laptops, it gets blurry. And for generative AI accessible through every smartphone on every train, it no longer suffices.
This article shows how Swiss universities are adapting their regulations: which new categories they introduce, which formulations have proven themselves, and which questions remain open.
The Four Types of Adaptation
In practice we observe four distinct strategies for updating regulations to the AI reality:
- Supplement: Existing exam regulations are supplemented with an AI article without changing the basic structure.
- Reformulation: The term "aids" is broadened so that generative AI is explicitly included.
- New category: AI is established as its own category alongside plagiarism and cheating, with its own rules and sanctions.
- Full rewrite: The exam regulation is fundamentally reworked with AI as a guiding theme.
For most Swiss institutions, a combination of Types 1 and 2 is practicable — a full rewrite is effortful and politically hard to push through, while pure supplementation often doesn't go far enough.
Case ETH Zurich: The Disciplinary Ordinance
ETH Zurich has adapted its disciplinary ordinance selectively. The central formulation on attempted cheating was extended to explicitly include "the use of generative artificial intelligence or comparable tools to create or substantially revise texts, insofar as this has not been explicitly permitted."
This formulation has several advantages:
- It names the topic clearly without citing individual product names (which would become outdated).
- It places the permission question at the center: what matters is whether use was allowed in the specific context — not whether it's allowed "in itself."
- It leaves room for flexibility at the course level. Faculty can regulate per module what is allowed and what is not.
Case EPFL: The Decentralized Approach
EPFL has barely touched its central exam regulation. Instead, course descriptions have been extended: every course leader must explicitly declare at the start of the course which AI use is allowed and which is not. These statements become part of the binding learning contract.
The advantage: maximum flexibility and context sensitivity. Programming courses can allow AI while a parallel closed-book exam takes place.
The disadvantage: responsibility shifts to individual lecturers. Without clear templates and training, a patchwork of varying rules emerges.
Case University of Zurich: The Taxonomic Solution
UZH has chosen the most comprehensive approach. Its policy distinguishes five levels of permitted AI use (from "no use allowed" to "free use") and requires each course to assign a level. The exam regulation references this and defines violations of the levels as attempted cheating.
This solution combines central steering with decentralized application. It is especially successful where faculties have the will to engage with the material. Where that will is absent, the system degenerates into formal box-checking without pedagogical depth.
The Legal Pressure Points
Regardless of the chosen approach, all regulation updates encounter the same legal pressure points:
Evidentiary Burden
Cheating must be proven, not merely suspected. Exam regulations that forbid AI use are only as effective as the institution's ability to document such use. Pure detector results often don't suffice in disciplinary proceedings — the regulation must clarify which evidence counts as sufficient.
A tested formulation: "Proof of attempted cheating may be established through technical verification, oral defense, analysis of the work process, or a combination of these means. None of these alone is conclusive."
Non-Retroactivity
Exam regulations do not generally apply retroactively. A tightening of rules cannot be applied to work submitted before the change. That's trivial — but occasionally overlooked in the rush of the AI debate.
Proportionality of Sanctions
Sanctions must be proportionate. Someone who writes an entire essay with ChatGPT acts differently than someone who had a paragraph smoothed. An exam regulation that sanctions both cases with expulsion is legally vulnerable. Tiered sanction models with different consequences by severity have proven themselves.
Right to Be Heard
Every sanctioned person has a right to be heard. Concretely: before any formal measure, they must receive the opportunity to respond to the accusation — orally or in writing. Exam regulations must anchor this right explicitly and regulate the process.
Practical Drafting Blocks for Swiss Universities
The following text blocks are tested and can be integrated into existing regulations.
Block 1: Extended Cheating Definition
Attempted cheating includes in particular:
a) submitting the work of others or parts of it as one's own (plagiarism);
b) using unauthorized aids in exams;
c) using generative artificial intelligence or comparable automated systems to create written work in whole or in part, insofar as this has not been explicitly permitted in the respective context;
d) false statements about one's own work process, in particular about the use of aids and tools.
Block 2: Declaration Obligation
With every written submission, students submit a declaration disclosing their use of generative artificial intelligence. The declaration is part of the submitted work. Knowingly false statements constitute attempted cheating under this regulation.
Block 3: Verification Means
The university reserves the right to verify written work with data protection compliant technical means for the use of generative artificial intelligence. The result of such verification alone does not establish an accusation of cheating but is part of a comprehensive assessment that in particular also includes conversation with the author, analysis of the work process, and stylistic comparison with prior work.
Block 4: Tiered Sanctions
For violations of the rules on AI use, the following sanctions are possible:
a) For minor violations: formal warning and condition to revise the work.
b) For substantial violations: grade of insufficient or grade deduction.
c) For serious or repeated violations: failing the examination, in especially serious cases exclusion from the course or program after prior hearing.
Sanctions depend on nature, scope, and context of the violation.
Block 5: Procedural Guarantees
Before any sanction under this regulation, the affected person receives the opportunity for written or oral response. They are informed of the grounds of suspicion and have the right to inspect the evidence used. Against sanctions from level b upward, appeal to the examination commission is permitted.
Open Questions Universities Haven't Yet Solved
Despite all progress, some questions remain largely unanswered:
How to Handle AI-Assisted Research?
Perplexity, Google NotebookLM, and similar tools research with AI but cite real sources. Is that research or AI use? Most exam regulations don't differentiate clearly — and thereby create uncertainty.
How to Handle Native Language Effects?
AI detectors have higher false positive rates on students whose first language is not the submission language. A fair exam regulation must account for this — but exactly how is open.
How to Handle Peer Tools Like DeepL Write or Grammarly?
These tools use AI for correction suggestions but are taken for granted in many academic contexts. Drawing the line from "real" generative AI is hard in practice.
How to Handle Student Co-Determination?
Switzerland has a strong tradition of student participation in university governance. Changes to exam regulations often require student approval. That slows processes but increases acceptance.
What Gymnasiums and Universities of Applied Sciences Can Learn From Universities
Universities have experimented a lot in the past three years. Gymnasiums and universities of applied sciences now adapting their regulations can draw lessons:
- Formulate flexibly: no product names, no rigid thresholds. Regulations must remain compatible with new tools.
- Levels instead of black-and-white: tiered permission and sanction models are fairer and legally more robust than yes/no rules.
- Procedural security: right to be heard, burden of proof, appeal rights must be explicitly regulated.
- Central tool selection: the exam regulation ideally references a centrally selected, compliant detector — avoiding fragmentation.
- Plan training: a new regulation without faculty training remains paper. Time and resources for training belong in the implementation plan.
The Timeline: When Should You Adapt?
The short answer: now, if you haven't already. The longer answer considers three factors:
- Legal certainty: without an updated exam regulation, sanctions are vulnerable to legal challenge.
- Pedagogical coherence: faculty need a framework within which to make their individual decisions.
- Institutional credibility: students take institutions more seriously when they actively shape developments rather than react to them.
The typical timeline for an update is three to six months: one month for a first draft, two months for consultation and feedback, one month for finalization and adoption, one to two months for communication before entry into force.
Conclusion: Regulations as Tools, Not Obstacles
A well-crafted exam regulation isn't a bureaucratic obstacle — it's a tool. It creates the framework within which faculty and students clearly know their roles, conflicts can be fairly negotiated, and sanctions — when they become necessary — are legally sound.
Swiss universities have gathered valuable experience over the past years. Institutions adapting their rules now can build on this experience and don't have to walk through the most painful learning curves themselves. That's the best news for all institutions still facing the update.
Sources
- ETH Zurich, Disciplinary Ordinance for Students, updated version 2024.
- EPFL, Course Guidelines on AI Use, 2024.
- University of Zurich, Academic Integrity Ordinance, 2024.
- swissuniversities, recommendations on exam regulations in the AI era, 2024.
- Federal Act on the Federal Institutes of Technology (ETH Act), SR 414.110.
- Cantonal university laws (ZH, BE, VD, GE as reference cantons).