Across Europe, institutions are moving assessment back towards the exam hall, and it is taking two forms. Some are returning to pen and paper: fixed seating, handwritten scripts, invigilators walking between the desks. Others keep the exam digital - students type on a PC - but bring it back into a supervised venue with traditional invigilation. The reasoning behind both is the same: generative AI has made it hard to know who wrote an unsupervised submission, and a supervised room is a familiar way to be certain again.
The pressure behind this is real, and it mostly comes from outside. Examiners raise concerns. Boards ask what is being done. Ministries and professional bodies want assurance that a degree still means what it says. When that pressure arrives, doing something visible matters - and returning to the room everyone remembers is the most visible thing available.
The instinct is understandable. But it is worth pausing on what the step back actually asks for - and on the fact that for a good number of these exams, it may not be necessary at all.
Both versions share the same operational bill, because both depend on the room. Rooms have to be found or hired at the same hour for large cohorts. Invigilators recruited, trained and rostered. Timetables rebuilt across departments, and again for the resit period. For an institution that has already digitalised, these are largely the costs and workloads that digitalisation removed in the first place.
Paper then adds a second layer. Processes designed around digital submission - anonymisation, distribution to markers, second marking, archiving - have to be rebuilt around physical scripts or a scanning step. And students are assessed, in part, on something the learning outcomes do not describe: handwriting. A cohort that has written digitally throughout its studies is suddenly measured on speed, stamina and legibility, and that does not fall evenly.
The digital-in-hall route avoids this. Students work in the environment they know, on a keyboard, and the marking workflow stays intact. But it brings its own bill, and one that is easily underestimated: several hundred machines in one space need power and cabling, wifi capacity that holds up when every candidate is connected at once, and a documented fallback for the moment connectivity drops mid-exam. Most venues were never built for that - least of all the large sports halls that get pressed into service precisely because they are the only rooms big enough. Retrofitting them costs real money, and it introduces a risk profile the paper exam never had: a technical failure that no longer affects one candidate but an entire cohort, in the middle of a summative assessment, with an appeals process to follow.
What neither route avoids is the room.
The room matters, because it is also where the fairness problem sits. Students with adjustments - extra time, rest breaks, a separate space, assistive technology, a quiet environment - are handled as exceptions in a venue: a room to find, a different clock to run, an assistive setup to reproduce away from the machine the student normally works on. This is a rapidly growing group. Across four Norwegian universities, applications for specially arranged exams more than doubled between 2020 and 2025, from 3,275 to 6,776 - and at the University of Oslo alone, applications rose 149% in the three years to 2025. Where the exam already runs on the student’s own device, most of that is configuration rather than coordination. Moving back turns it into a coordination task again — and the students least able to absorb that are the ones who carry it.
The practical costs show up in a budget. The strategic cost does not. Going back to paper means switching the technology off rather than learning to use it; the digital-in-hall route does not go that far, but it does re-attach digital assessment to the building it had been freed from. Either way the formats stay as they were instead of being developed and differentiated, which pauses work many institutions have spent a decade on: assessment that looks more like the tasks graduates will actually face, in the tools their disciplines actually run on.
Going back has the appeal of a warm room. But it is the warmth of a fire nobody is refuelling. Generative AI is not going away, and neither is digital assessment. A response that depends on switching both off buys quiet for a cycle or two, and leaves the same question waiting - with less capability in the building to answer it.
This is clearest for the institutions choosing the digital-in-hall route. They have already accepted that the exam can be digital. What the venue is actually buying them is supervision - and supervision can be added to the exam the student already sits, on the device they already use, at home or on campus. For them, the remaining difference between the two options is the room itself, and the room is where nearly all of the cost and most of the inequity is concentrated.
That is what proctoring does, and its practical appeal is that very little else has to change. The exam format stays. The digital workflow stays - anonymisation, distribution, marking, archiving. Open-book stays possible, and so does letting students work in Excel, an IDE, a CAD package or a statistical environment. No rooms, no rebuilt timetable, and none of the venue infrastructure described above.
Two things need examining properly, because they are where proctoring projects stall.
The first is cost, and here many institutions are working from a number that is five years old. Proctoring entered most people’s awareness during the pandemic: procured at speed, under emergency conditions, and largely built on live human invigilation of every candidate. It was expensive, and the memory of that price has since hardened into an assumption.
A good deal has changed. Automated review now handles the routine session and escalates only what needs a human eye - AI in the mix as a way of flagging what a person should look at, not as a replacement for the person. That is what turns supervision from a per-sitting expense into something an institution can apply across a portfolio rather than ration to a handful of finals. Proctoring today can be affordable in breadth, for the many rather than the few. Whether it is depends entirely on the provider and the commercial model behind it - which is why the figure worth working from is a current quote, not a pandemic-era recollection.
The second is where the data goes. A significant part of the proctoring market processes recordings, biometric identity data and behavioural analysis outside the EU, leaving a European institution to defend a transfer mechanism it did not choose - often with a DPIA written after procurement rather than before it. For many assessment and legal teams this, more than price, is why a proctoring project never leaves committee. Neither problem is inherent to proctoring; both are procurement questions, and both are answerable.
On how supervision is actually experienced, the evidence worth reading comes from Norway. Writing in Khrono in August, Oslo Nye Høgskole rector Sander Sværi, Abelia’s Øystein E. Søreide and NTNU master’s student Kjersti Nilsson Hannestad describe what happened when ONH piloted digital invigilation - camera, audio and screen sharing, with the video analysed automatically and flagged deviations checked by hand. Hannestad, a former student representative at the institution, reports that the exam proved less intrusive than students had feared, and the grade distribution returned to where it sat before the AI tools arrived. Their conclusion is that the fight against AI cheating will not be won in the gymnasium.
That reflects something we hear often. Supervision applied through software - whether reviewed automatically or by a person - is frequently experienced as more neutral than supervision in a hall. It treats every candidate identically, it is documented, and a student who disputes a flag has a record to point to. The condition is that the system flags and a person decides. It is also why proctoring belongs as one instrument among several rather than a policy for everything - alongside oral defences, staged and supervised drafting, and physical exams where they genuinely fit the learning outcomes. It suits exams where authorship and identity matter, and it should be off where they do not.
WISEflow live proctoring adds real-time remote invigilation to an exam a student sits on their own device: identity verification, webcam, screen and audio recording, and automated flagging, reviewed live by a human proctor or after the exam. It sits at the top of four proctoring levels in WISEflow, alongside device monitoring for open-book and tool-based exams, the lockdown browser, and light identity assurance using periodic stills rather than continuous video.
The levels exist so that supervision can be set per exam rather than by institution-wide decree - proportionate to what is actually at stake. That only works in practice if the commercial model allows it, which is why ours is built for supervision to be applied where an exam needs it rather than rationed to the few sittings a budget can absorb. And the data stays in the EU by default, on European infrastructure, retained to a defined schedule and covered by feature-specific DPIA documentation - not routed abroad and reconciled afterwards. That distinction matters more each year as the EU AI Act’s obligations come into force. We wrote about the layering itself in The closed-book exam comeback.
Concern about AI in assessment is legitimate, and some exams genuinely belong in a supervised room. The question is whether the whole portfolio does, and whether an institution wants to answer a technology problem by removing technology.
Our work with universities is about the other direction: adapting assessment formats to what is now possible, differentiating supervision by what each exam actually requires, and keeping the capability that has been built. That is slower than clearing the hall for June. It also still works next year.
Before committing to a rollback, cost it - rooms, invigilators, cabling and network, redesigned processes, and the students it puts at a disadvantage. Paper carries all of it; a digital exam in a supervised venue carries most of it.
Then check which of those exams could simply be supervised where the student already sits - and price that properly, against a provider whose model does not make scale the problem and whose data stays in Europe.
The aim is to develop assessment formats, not to dismantle them.
Sander Sværi, Øystein E. Søreide and Kjersti Nilsson Hannestad, “Kampen mot KI-fusk vinnes ikke i gymsalen”, Khrono, 25 August 2026.
Applications for specially arranged exams at Norwegian universities, 2020–2025: Universitetsavisa, 1 June 2026; see also Universitas, 25 November 2025.