Goodbye Homework: How Exams Are Changing in the AI Era 2026
How exams are changing in the AI era 2026: universities replace homework with oral defenses, in-class writing, and process portfolios as ChatGPT reshapes assess
Universities across the United States, the United Kingdom, and Australia are quietly retiring the traditional take-home essay. According to a survey of 1,396 instructors published by Tyton Partners in March 2026, 61% of higher-education faculty have already redesigned at least one assessment to account for generative AI, and 38% have eliminated unsupervised written homework altogether. The shift, accelerated by the release of GPT-5 and Claude Opus 4.7 over the past eighteen months, is being framed by deans as the most significant change in how exams are changing in the AI era since the introduction of standardized testing.
For students entering the 2026-27 academic year, the implications are concrete: more oral defenses, more in-class handwritten work, more process portfolios with version history, and fewer evenings spent polishing a 2,000-word essay at home. The question administrators are now asking is not whether AI can write the assignment, but whether the assignment was ever measuring what they thought it measured.
- A Tyton Partners survey from March 2026 found 61% of US faculty have redesigned at least one assessment because of generative AI.
- The University of Sydney rolled out a two-lane assessment model in February 2026 that splits coursework into secured and AI-permitted tracks.
- UNESCO reported in its 2025 Global Education Monitoring update that only 21% of countries have national guidelines on AI use in classrooms.
- Oral defenses and in-class writing are returning as the dominant high-stakes formats in humanities, law, and business programs.
Why traditional homework is collapsing
The take-home essay is collapsing because it can no longer reliably distinguish a student’s reasoning from a model’s output. A January 2026 study from Stanford’s Human-Centered AI Institute found that experienced graders correctly identified AI-written undergraduate essays only 53% of the time, barely above chance, even when using commercial detectors.
That figure, replicated in similar trials at the University of Cambridge and the University of Toronto, has effectively ended the detector arms race. Turnitin’s own AI-writing indicator, once treated as forensic evidence, is now described in the company’s April 2026 guidance as «an investigative signal, not proof.» Several US universities, including Vanderbilt and Michigan State, have disabled the feature entirely, citing false-positive rates that disproportionately flagged non-native English speakers.
The result is a credibility problem. If graders cannot tell whether a paper was written by a student, and detectors cannot either, the grade itself loses meaning. Faculty senates from Berkeley to Edinburgh have spent the past academic year debating the same uncomfortable conclusion: assessments that happen outside the classroom now measure access to AI, not mastery of the subject.
The new exam formats replacing homework
Four formats are emerging as the dominant replacements for unsupervised written homework: in-class handwritten essays, oral vivas, process portfolios with timestamped drafts, and AI-collaborative tasks where the use of models is explicit and graded. Each addresses a different weakness in the legacy take-home model.
The University of Sydney’s two-lane policy, announced in February 2026, has become the most-cited template. Lane one covers «secured» assessments — supervised, in-person, no AI permitted — and is used for grading. Lane two covers «open» tasks where students must use AI and document their prompts, critiques, and revisions. Both lanes count toward the final mark, but only lane one determines progression.
Oral defenses, long standard in continental European doctoral programs, are spreading downward. At Imperial College London, second-year computer science students now defend their coursework in fifteen-minute panels. At Bocconi in Milan, undergraduate economics seminars end with a viva instead of a paper. The format is expensive in faculty hours, but administrators argue it is the only one that scales with model capability.
The handwritten comeback
Blue books, the bound paper notebooks once associated with mid-twentieth-century college life, are selling at record volumes. Texas A&M’s campus store reported a 480% increase in blue book sales between September 2024 and September 2025, according to figures shared with The Wall Street Journal. Similar spikes have been documented at Berkeley and the University of Florida.
Handwriting is not pedagogically superior on its own merits — research from the University of Stavanger has long suggested some cognitive benefits, but the case is contested. The appeal in 2026 is pragmatic: a closed room, a paper booklet, and a proctor remain the cheapest verified-human assessment available.
What students and instructors are saying
Instructor reactions split along disciplinary lines. Humanities and law faculty report the steepest redesigns, while STEM programs, which already relied on supervised exams and problem sets, report comparatively minor changes. Students, polled by the Higher Education Policy Institute in the United Kingdom in November 2025, describe the new formats as fairer but more stressful.
«We are not banning AI. We are designing assessments that survive its existence. The take-home essay was already a flawed instrument; the model just made the flaws impossible to ignore.»
Students interviewed by the Chronicle of Higher Education in March 2026 described mixed feelings. Many welcomed the clarity of supervised exams over the ambiguity of policies that warned against AI use without defining it. Others worried that oral defenses penalize quieter students and those with English as a second language, a concern echoed in early equity audits from the Open University.
The professional exam world is moving in the same direction. Candidates preparing for the Spanish Medical Residency Exam 2026 already use AI tutors for mock tests at home, but the exam itself remains rigorously supervised and on paper. That split — AI-saturated preparation, AI-free verification — is the model most universities are now copying.
The data picture: how exams are changing in the AI era worldwide
Beyond the Anglophone world, adoption of new assessment models is uneven. UNESCO’s 2025 Global Education Monitoring update reported that only 21% of surveyed countries have issued national guidelines on AI use in classrooms, and fewer than 10% have updated their high-stakes examination protocols. The gap between institutional response and national policy is one of the defining tensions of how exams are changing in the AI era.
| Format | AI-resistance | Faculty cost | Equity concerns |
|---|---|---|---|
| In-class handwritten exam | High | Low | Disadvantages students with dysgraphia or motor impairments. |
| Oral defense / viva | Very high | High | Penalizes shy speakers and non-native English speakers. |
| Process portfolio with version history | Medium | Medium | Requires access to monitored platforms. |
| AI-collaborative task | Low (by design) | Medium | Depends on tool licensing access. |
The commercial EdTech sector is adjusting accordingly. Established players such as Quizlet and Duolingo have launched proctored study modes, while smaller startups like Modo Cheto or Memrise emphasize self-paced practice that mirrors the new exam formats. The center of gravity is shifting from content delivery to verified practice — drills that look like the supervised exam students will eventually sit.
Implications for students and the sector
For students, the practical consequence is that study habits built around polishing take-home work no longer map to how they will be graded. Programs that emphasize active recall, timed writing, and articulating arguments out loud are becoming more valuable than those that optimize a single submitted document. Universities that delay the redesign risk awarding degrees that employers begin to discount.
Employers are already signaling preferences. A survey of 312 hiring managers by the Burning Glass Institute in April 2026 found that 44% now place greater weight on supervised certifications and live technical interviews when AI use in coursework is suspected. The signal value of an unverified transcript is eroding, especially for entry-level knowledge work.
Universities themselves face a structural cost. Oral defenses and supervised exams require more faculty hours, more rooms, more proctors. Smaller institutions without endowment cushions are likely to feel the squeeze first, raising the prospect of a two-tier system where well-resourced campuses can afford verified assessment and others cannot. Accreditation bodies have begun asking pointed questions about this gap.
What comes next
The 2026-27 academic year will be the first in which most flagship universities enter the term with formal AI-era assessment policies in place rather than emergency memos. Whether those policies survive contact with the next generation of models — multimodal agents capable of sitting interactive exams alongside students — is the open question for 2027.
For now, the trajectory is clear: less homework, more proctoring, and a renewed argument over what a degree is supposed to certify. The blue book is back, the viva is spreading, and the take-home essay is becoming a historical artifact. The harder conversation, about whether universities ever truly measured thinking or only its written residue, has barely begun.
