Large language models have broken the assumptions underlying our most common assessment formats. A take-home essay, a programming assignment, a reflective journal, each used to carry an implicit guarantee that the work was produced by the student. That guarantee is gone, and “AI-detection” tools are too unreliable to restore it. At the same time, KAU’s examination regulations (RB 68/19, Ch. 6) require an individual assessment whenever group work contributes to a grade. We are caught between a regulatory requirement and a technological reality, and most of the obvious responses such as, banning AI, banning group work, scaling oral vivas across 130-student cohorts, either fail at scale or undermine the pedagogy we want to keep. This presentation offers a small, tested response: a diagnostic framework, and three concrete design moves I now use across my own courses. The framework asks three questions of every assessment:
- Can it be outsourced (to AI, peers, or paid services)?
- Does it scale across the cohort sizes we actually teach?
- Does it cover the CLO it claims to test?
If any of the three fails, the assessment is broken, even if it looks defensible on paper. From this framework I will walk through three design moves: locked-environment practical examinations (Enterprise Database Applications, Higher Colleges of Technology, UAE – adopted across 16 campuses); a U/G group component paired with a U/G/VG written exam whose questions can only be answered by a student who genuinely contributed to the project (ISGA03 and ISGB36 at KAU); and in-class oral check-ins built into existing tutorial sessions as a continuous, lightweight individual signal. Together these provide a defensible audit trail for grade decisions without requiring 32 hours of viva contact per assessment cycle. I will also be open about what does not yet work. Oral vivas remain infeasible at first-cycle scale. Open-book online exams shift the design problem rather than solving it. Practical examinations depend on infrastructure (locked-down computers, exam-hall capacity) that we do not always have. The intent is not to claim a solution but to share a workable practice, and to invite the room to push back on it. - The problem (3 min): What take-home assessments used to verify; what LLMs
broke; the regulatory gap between group submissions and individual accountability. - A diagnostic framework: Three questions before choosing any assessment format.
- Three concrete responses: Locked-environment practical exam (HCT / Oracle
Apex); U/G group + discriminating written exam (ISGA03, ISGB36); tutorial oral
check-ins. - What does not yet work: Honest account of oral-viva scaling, open-book exams, AIdetection unreliability, and infrastructure constraints.
- Discussion: What is the minimum viable individual assessment in your course?
Relevance for KauKan
The problem is relevant across every department at KAU. Anyone who currently uses takehome work, group projects, or reflective writing is facing the same regulatory and technological pressure. The talk offers a vocabulary and a practice that participants can apply to their own course, regardless of discipline.
