96% at home, 48.6% in the classroom: the gap that pushes universities to rewrite their exams

Flipped lectures, handwritten essays in three hours, defensive orals, screen-free rooms: faced with human correctors unable to spot AI, campuses are moving evaluation under surveillance.

The figure has been circulating in American economics departments: at Brown, Professor Roberto Serrano saw the average for a homework assignment reach 96%, while that of the final exam taken in the classroom fell to 48.6%. The gap sums up what universities are facing at the start of the 2026 academic year.

The accompanying surveys are equally telling. Nearly 90 percent of Harvard undergraduates say they use AI for their academic work, and about a quarter use it instead of assigned readings. A study by the University of Reading also showed that human proofreaders identified AI-produced copies in only 6 percent of cases—condemning any strategy based on detection.

Moving work rather than prohibiting it

The dominant response is to bring intellectual output back to the classroom. Taiyo Inoue at Cal State Maritime Academy has switched her lectures online to devote the sessions to problem-solving on the blackboard. At Calvin University, Katie Day Good has analytical essays written by hand, in three hours on time. The University of Chicago is testing classrooms without electronic devices in first-year law school. The University of Sydney, where only one in eight students still attends lectures, has adopted a "two-track" assessment model that distinguishes between what can be produced with AI and what needs to be defended in person.

Others turn the tool upside down. At NYU Stern, Panos Ipeirotis built Viva, an AI-driven oral examiner that adapts the difficulty of its questions to the student's answers. The common logic remains the same: accept that the submitted assignment no longer proves anything by itself, and re-establish a moment when the student must answer for his or her work.