The essay was perfect. The student could not explain a word of it. That moment — increasingly familiar to professors at universities across Europe and North America — is not just a cheating problem. It is a signal that the assessment model built around take-home work has structurally failed. AI did not break education. It exposed how much of education was already running on a system that trusted outputs over understanding.
The Collapse of Take-Home Authorship
Since the public release of ChatGPT in late 2022, educators have been grappling with a rapidly evolving challenge. Assignments that once required hours of research, synthesis, and original thought can now be completed in minutes. Take-home exams, reflective essays, coding assignments, and even problem sets have all become vulnerable to what experts describe as “cognitive offloading” — where students outsource thinking to machines.
The scale is significant. An Inside Higher Ed–College Pulse survey of US college students found that 85% had used AI in their coursework. In the UK, Russell Group universities including UCL and Imperial have recorded sharp increases in AI-related academic misconduct cases — though detection remains inconsistent and contested. AI detection tools offer limited reassurance, with accuracy ranging between 33% and 81% depending on the tool. The authorship of a submitted piece of work is, for practical purposes, no longer verifiable.
This is the verification crisis at the centre of the current debate. It is not primarily about dishonest students. It is about an assessment architecture built on the assumption that a submitted essay reflects independent thought — an assumption that no longer holds.
Why Detection Cannot Solve the Problem
The instinct to fight AI with AI detection tools has largely failed. Attempts to combat AI misuse through detection software have largely failed, with accuracy rates varying widely and false positives creating their own problems. Universities that have tried to ban AI tools entirely face an enforcement challenge that is effectively impossible at scale. And the boundary between legitimate AI assistance and full substitution is genuinely unclear — for students and institutions alike.
The deeper issue is that the problem is not technological. It is architectural. A 2025 study conducted by academic publisher Taylor & Francis found that generative AI use in higher education has created a “wicked problem” that will require a diverse set of solutions. No single tool or policy resolves it. What is required is a redesign of what assessment is measuring — and that is a much harder task.
The Return of the Oral Exam
The response that has gained most traction is not technological. It is ancient. “You won’t be able to AI your way through an oral exam,” says Chris Schaffer, a biomedical engineering professor at Cornell who introduced an oral defence format last semester. Penn’s Center for Teaching and Learning has described a “massive shift toward in-person assessments.” The university is among a growing number running faculty workshops specifically on oral exam design.
The oral exam predates the university as an institution. In European traditions — the Oxbridge tutorial, the German Kolloquium, the French grand oral — spoken assessment has always been part of the educational architecture. In the US, it largely disappeared from undergraduate education. It is now returning, not as nostalgia, but as the most reliable method of verifying whether a student actually understands the material they submitted.
At NYU, several types of oral assessments are on the rise — office hours requirements, presentations, cold-calling in class. “I need to look my students in the eye and ask, ‘Do you know this material?'” says Clay Shirky, vice provost for AI and technology in education. Stanford’s SCALE project has proposed scalable conversational exam formats — structured oral assessments designed to work at undergraduate volume, evaluating not answers but the ability to reason through and explain them.
From Answers to Reasoning
The shift the oral exam represents is not just logistical. It is conceptual. The take-home essay measured a product. The oral exam measures a process. That distinction is, in the AI era, the only one that reliably tracks actual learning.
This phenomenon threatens the foundational purpose of education itself: developing reasoning, judgment, and intellectual endurance. A student who submits an AI-generated essay and cannot explain it has not learned to write, research, or argue. They have learned to commission. Those are different skills — and the second is not what a degree is supposed to certify.
As explored in “When Systems Decide for Us, What Happens to Human Judgment?,” the deeper question is not whether AI assistance is legitimate but whether the humans using it retain the capacity to evaluate, interrogate, and extend what it produces. The oral exam is one answer to that question. It forces the student to demonstrate that the thinking happened — not just the output.
Homework as a Different Kind of Task
None of this means homework disappears. It means its purpose changes. Faculty members are no longer confident that submitted work reflects a student’s understanding. Students navigate an environment where the boundaries between acceptable assistance and outright substitution remain unclear and inconsistently enforced.
The assignment that survives the AI era is one designed around reasoning rather than production — where the submitted work is a starting point for a conversation, not a finished object to be graded in isolation. Process portfolios, iterative drafts with reflections, annotated bibliographies that require the student to evaluate sources: these are formats that AI can assist with but cannot substitute for entirely.
As explored in “Education Is Becoming a Subscription Service,” the broader transformation of education is already underway. The specific question of homework and assessment is one part of that shift — but perhaps the most immediate, because it touches every student, every week, right now. AI did not kill homework. It forced a reckoning with what homework was actually for.
Key Sources
- eWeek, “Colleges Turn to Oral and Handwritten Exams as AI Disrupts Assessments”: https://www.eweek.com/news/colleges-turn-to-oral-exams-ai-disruption/
- Fortune, “The Gen Z stare meets the mysterious perfect homework assignment”: https://fortune.com/2026/03/25/oral-exams-colleges-anti-ai-teaching-method-gen-z-stare/
- AP / AOL, “Perfect homework, blank stares: Why colleges are turning to oral exams”: https://www.aol.com/articles/perfect-homework-blank-stares-why-040137824.html
- Washington Post, “College professors use oral exams to combat AI”: https://www.washingtonpost.com/education/2025/12/12/ai-artificial-intelligence-college-oral-exam/
- Stanford SCALE, “Conversational Exam: Scalable Assessment Design for the AI Era”: https://scale.stanford.edu/ai/repository/conversational-exam-scalable-assessment-design-ai-era
- Business Insider, “AI college exams wicked problem”: https://www.businessinsider.com/ai-college-exams-wicked-problem-no-clear-fix-researchers-warn-2025-9
Subscribe to EuroLuminant for independent European journalism.
