Office for Students says AI lets some students finish university assessments in minutes
The regulator wants more closed-book exams, presentations and practical tests, and says it will act where universities fall short.
England's higher education regulator wants universities to make far more use of exams and other assessments that AI cannot do for students, after finding that AI tools can complete many unsupervised assignments to a high standard.
Research published on Wednesday by the Office for Students (OfS) and Advance HE, a charity that promotes university teaching, found that a large share of assessments are still done without supervision. Students told researchers about classmates who had finished assessments with chatbots, sometimes within a few minutes, and been given similar marks to those who did the work themselves. The report said it was often possible to use AI in ways staff could not reliably detect.
The findings are based on interviews and discussions with more than 300 staff and students at 127 universities and colleges. Many students said they found their university's rules on AI hard to understand, and some felt staff use of AI had made teaching materials worse. They also saw AI skills as important for getting a job and wanted more chances to learn them.
What the regulator wants
The OfS said formats designed to keep AI out, such as closed-book exams, presentations, practical assessments and performances, were in many cases likely to give the most accurate measure of what a student has learned. It will require universities to make sure their assessments are valid and reliable, which it said will often mean secure, AI-free tests.
Where it finds a university falling short, the regulator can impose specific conditions of registration, issue fines or suspend access to student support and OfS funding.
How universities are responding
Many have already changed how they assess students, bringing in more in-person exams and vivas, which are spoken examinations.
Advance HE plans to start a programme early in 2027 to help university leaders set out how their institutions will use AI.
