England's university regulator calls for more assessment without AI and reopens the debate on how to verify authorship
The Office for Students recommends secure formats such as in-person or oral exams. In the group, members debated what can and cannot be detected.
The Office for Students (OfS), the higher education regulator in England, published its page on artificial intelligence in higher education on 11 November, drawing on work done with Advance HE. The report, Making AI work for students, warns that many unsupervised tasks can now be completed with AI to a high standard and with little involvement from the student, and calls for assessments to measure what students know and can do, often through secure, AI-free formats: in-person exams, presentations, oral exams and practical tasks. It acknowledges that these formats have costs and may disadvantage some disabled students, and it also describes assessments in which AI use is permitted or required.
The news led the group into a deeper discussion. One view holds that short online courses have lost their point, because an assistant with a browser extension can complete them on its own. Another recalls that, as mentioned in the group, there are Moodle plug-ins that try to detect AI from the speed or regularity of answers, but that there are also extensions that write like a person and prompts that make text undetectable. From there two conclusions emerge that sit side by side: that accusing someone of using AI is not the way forward, and that, in an electronic environment, any parameter of what is human can be imitated if a person can do it.
The report does not consider detectors reliable either: it is difficult to establish conclusively whether AI was used, and a detector's output should be, at most, one more piece of evidence, with suspicion never lowering a grade. The paper has already covered the debate on detectors and the “witch hunt” and how a detector percentage does not prove copying.
Note added on 13 November 2026: the piece took from Unite.AI's coverage what the Office for Students and Advance HE report says. It now cites the report directly, and two sentences have been adjusted to its wording.
England's university regulator calls for more assessment without AI and reopens the debate on how to verify authorship
The Office for Students recommends secure formats such as in-person or oral exams. In the group, members debated what can and cannot be detected.
The Office for Students (OfS), the higher education regulator in England, published its page on artificial intelligence in higher education on 11 November, drawing on work done with Advance HE. The report, Making AI work for students, warns that many unsupervised tasks can now be completed with AI to a high standard and with little involvement from the student, and calls for assessments to measure what students know and can do, often through secure, AI-free formats: in-person exams, presentations, oral exams and practical tasks. It acknowledges that these formats have costs and may disadvantage some disabled students, and it also describes assessments in which AI use is permitted or required.
The news led the group into a deeper discussion. One view holds that short online courses have lost their point, because an assistant with a browser extension can complete them on its own. Another recalls that, as mentioned in the group, there are Moodle plug-ins that try to detect AI from the speed or regularity of answers, but that there are also extensions that write like a person and prompts that make text undetectable. From there two conclusions emerge that sit side by side: that accusing someone of using AI is not the way forward, and that, in an electronic environment, any parameter of what is human can be imitated if a person can do it.
The report does not consider detectors reliable either: it is difficult to establish conclusively whether AI was used, and a detector's output should be, at most, one more piece of evidence, with suspicion never lowering a grade. The paper has already covered the debate on detectors and the “witch hunt” and how a detector percentage does not prove copying.
Note added on 13 November 2026: the piece took from Unite.AI's coverage what the Office for Students and Advance HE report says. It now cites the report directly, and two sentences have been adjusted to its wording.
