The Office for Students says AI can let students complete some university assessments to a high standard in minutes, sometimes without reliable detection. The regulator’s warning raises a practical challenge for universities: assessments need to show what students know, not merely what a chatbot can produce.
Watch Desk analysis
What happened
A study by the OfS and Advance HE gathered views from 244 higher-education staff and 91 students at at least 127 universities, the BBC reports. It found that a significant proportion of assessments take place in unsupervised formats where students can access AI, and that it is frequently possible to use AI in ways that cannot be reliably detected.
The OfS said closed-book and practical exams were the most effective assessment formats. Its director of strategy and delivery called for far greater use of AI-free assessment, while the regulator’s research also recommends replacing an assessment if AI can complete it without testing the intended knowledge and skills. The BBC reports that some staff had encountered assignments that looked alike and questioned whether their institution’s assessments remained fit for purpose.
Universities UK says many institutions have already changed their approach, including more in-person exams, presentations, vivas, process-based assessment and controlled uses of AI. The BBC also reports that OfS proposals under consultation could restrict tuition-fee increases or impose student-number caps on universities that fail to meet quality standards.
Why it matters
This is not a claim that every student is using AI to cheat. It is a warning about assessment design, supported by a study of staff and student views rather than a count of how often AI use occurs across all university work. The distinction matters: the problem described is that some assessments may no longer measure the skills they are meant to test.
Universities face a choice beyond trying to detect AI-written work after the fact. They can use more supervised or practical assessment, redesign coursework around process and judgement, or make controlled AI use part of the task. Different subjects will need different answers; an exam hall is not a universal substitute for good teaching.
Our read
The most useful advice is also the least glamorous: if a task can be completed by AI without demonstrating the learning it is supposed to assess, replace the task. Detection tools may help in some cases, but they cannot rescue an assessment that tests the wrong thing. Universities should explain clearly when AI is allowed, and make sure students can show their own understanding where it matters.
What to watch
- Whether universities publish concrete changes to assessment formats and their results.
- How the OfS’s proposals on tuition-fee increases and student-number caps develop through consultation.
- Whether institutions make AI rules clearer to students and provide staff with the support to apply them.
Discussion spark: Should universities move more assessment into supervised exams, or redesign coursework so students can use AI while still proving what they know?
Sources and evidence
- Source update (11 November 2026, 00:21 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.