Too many university students in England can use AI to complete assessments without staff finding out, the higher education regulator has said.
The Office for Students (OfS) found that "a significant proportion" of assessments are unsupervised, and that many can be completed to a high standard with AI tools — sometimes in minutes.
The sentence that does the damage
The report concludes it is "frequently possible" for students to use AI "in ways that cannot be reliably detected".
That is the finding, and it is a different claim from the one usually made. The familiar worry is that students are cheating. This is a statement about the institution: that the method universities rely on to tell good work from generated work does not function.
The OfS says this poses a "significant risk" to the credibility of those assessments. Note the object of the sentence — the risk is to the qualification, not to the student.
What the research is
The work was carried out by the OfS with Advance HE, a charity that gives awards for teaching excellence. It is based on the views of 244 higher education staff and 91 students from at least 127 universities.
It is worth being accurate about what that is: a survey of opinion across a wide spread of institutions, not a measurement of how many students actually used AI. It tells you what staff and students believe is happening and what formats are vulnerable. It does not tell you the rate.
Where the exposure is
Unsupervised assessment includes coursework, but also remote online exams. Those were rolled out widely during Covid and never fully withdrawn — research from Swansea University last year suggested 93 UK universities still use them in some form.
The OfS says closed-book and practical exams are the most effective formats. That is a return to the oldest answer available: if you cannot verify the conditions a piece of work was produced under, control the conditions instead.
What happens to universities that do not move
Universities UK said many institutions had "made substantial changes" to make assessments "AI-proof".
There is a consequence attached. Under proposals now out for consultation, universities that fail to meet OfS quality standards could be banned from charging increased tuition fees, or have caps imposed on their student numbers.
Why this reaches Dhaka
Bangladeshi universities run the same formats — take-home assignments, unsupervised online components left over from the pandemic — and have the same detection problem, without a regulator publishing research about it.
The practical lesson is the one the OfS reached: detection tools are not the answer, because the report's central finding is that detection fails. Assessment design is. That is slower and more expensive, and it is the only part of this within a department's control.
What closed-book exams actually cost
The regulator's preferred answer is not free, and it is worth naming the price rather than pretending the fix is simple.
Supervised exams need halls, invigilators and fixed dates. They are harder for students with caring responsibilities, for part-time students, and for anyone whose disability is better accommodated by extended coursework. Universities moved away from them for reasons, not only because it was cheaper.
So the choice is not between honest assessment and dishonest assessment. It is between a format that can be verified and one that is accessible, and institutions are being asked to buy the first with some of the second.
