England's regulator says AI has made coursework hard to judge fairly
England's higher education regulator urges universities to return to traditional exams and in-person presentations as AI complicates grading.
According to the Guardian, England's Office for Students has called on universities to go back to traditional exams and in-person presentations when grading students. The regulator's investigation found that lecturers say artificial intelligence has made it "close to impossible" to assess coursework reliably.
The Guardian reports that the Office for Students warned the credibility of degrees awarded by English universities is under threat. Its concern centers on the widespread use of AI tools in essays, dissertations and open book exams, areas where a student's own work has traditionally been judged.
The outlet also reports that students have described peers boasting about using AI without being detected by staff. That account, as reported, points to a gap between what students are doing and what assessors can confirm.
The Guardian's report presents the regulator's recommendation as a response to these concerns rather than a finished set of rules. The reporting does not spell out how quickly universities would be expected to change their assessment methods, or what enforcement, if any, would follow.
The findings land at a time when many institutions are still working out how to handle generative AI in teaching. The regulator's position is a clear signal that the sector's current approach is not seen as sufficient by at least one major oversight body.
Background
Universities have long relied on a mix of coursework, essays, dissertations, labs and timed exams to test what students know. Coursework became the dominant form of assessment in many subjects partly because it allows deeper research and writing than a short exam can. It also depends on an assumption that the submitted work reflects the student's own effort and understanding.
Plagiarism has always been a concern, and institutions developed tools and policies to deal with copied text, contract cheating and collusion. Text-matching software helped detect copying from known sources, but it was less effective at spotting original text that had been generated on demand.
Generative AI systems changed that picture quickly. Tools that can produce fluent essays, summaries and code are widely available, and they can be prompted to match a particular style or word count. Detection tools that claim to identify machine-written text have been controversial, with questions raised about their accuracy and about the risk of wrongly accusing students.
Many universities responded with a mix of approaches. Some revised honor codes and disclosure rules, some redesigned assignments to focus on process, drafts and oral discussion, and some reintroduced supervised assessment. Others have struggled to agree on a single policy across departments. Assessment design is a slow and costly process, which is one reason the debate has been difficult to resolve.
Why it matters
The question here goes beyond cheating. A degree is a signal to employers, graduate schools and the public that a holder has reached a certain level of knowledge and skill. If assessment cannot reliably separate a student's own work from machine output, that signal weakens, and the value of the qualification can be questioned.
The issue also affects fairness. Students who follow the rules may feel disadvantaged if others use AI undetected, and staff may feel pressure to police work they have little practical means to verify. Trust between students, lecturers and institutions can erode when assessment feels like guesswork.
There is a real tension, too. Traditional exams and in-person presentations test recall and performance under conditions that differ from how many professionals work, where tools are routinely used. Returning to them may protect the reliability of grades while raising questions about what skills are being measured and whether some students are disadvantaged by time pressure or anxiety.
The regulator's stance also matters because it comes from an oversight body with influence over the sector. Even when guidance is not binding in every detail, it can shape institutional decisions, accreditation conversations and public expectations about what a degree from England should guarantee.
What to watch
Watch how universities respond to the recommendation. Some may adopt more supervised assessment, while others may argue for redesigned coursework that asks students to show drafts, reasoning or oral defenses of their work. The practical cost of either path will be a major factor.
Pay attention to whether the regulator sets out specific expectations, timelines or monitoring. The Guardian's report does not indicate how any change would be enforced, so the next official statements will be important.
Also watch student and staff reactions, as well as evidence on whether new assessment methods actually reduce undetected AI use without creating new unfairness.
Viewpoints
Commentary from our columns: three angles on this story.
The Skeptic open questions and weak points
It is easy to hear "return to exams" and feel that the problem has been solved. It has not. Exams carry their own weaknesses. They reward fast recall, penalize students with certain disabilities or test anxiety, and say little about whether someone can sustain a long research project. Swapping one imperfect method for another is not a fix.
There are also open questions the headline skips. How widespread is undetected AI use, really? The report describes student claims about peers, and anecdotes about boasting are not the same as measured data. Institutions should be cautious about designing whole systems around a problem whose size is still unclear.
Before universities rush to change, they should ask what they are actually trying to measure. If the goal is to confirm independent understanding, the answer may be a mix of methods rather than a single return to the past. Good skepticism here means demanding evidence, not simply choosing the most familiar option.
The Builder the engineer's view
From an engineering standpoint, the core problem is that most assessment was built on an assumption that is no longer safe: that a finished text is evidence of the process behind it. Detection tools that try to infer authorship from style are fragile, and any system that relies on them will produce false positives and false negatives.
A more durable design treats assessment as a pipeline. Drafts with timestamps, version history, short oral checkpoints and tasks tied to specific class material all create signals that are harder to fake at scale. None of this is perfect, but layered evidence is more robust than a single submission.
The regulator's suggestion of in-person work is one valid layer. The challenge for universities is to build assessment that is observable without becoming so burdensome that staff cannot run it. Good tooling and clear process design will matter as much as policy language.
The User what changes for people who use it
For students, the biggest change is likely to be felt in how they prepare, not just how they are graded. If essays and take-home work are no longer the main test, study habits shift toward memorization, practice under timed conditions and speaking through ideas out loud. Some students will welcome that clarity. Others will find it a harder route to demonstrate what they know.
Lecturers will feel it too. Marking loads may change, and the rhythm of a course could shift toward more in-person checks. Those changes sound manageable on paper, but they take time, rooms and staff, which not every department has in abundance.
For the students who have been doing the work honestly, the most important question is whether the new system will recognize their effort. A fair assessment should reward understanding, not just the ability to perform under a clock. That is the standard universities should hold themselves to as they adjust.