Academic misconduct cases involving AI at Edinburgh University went from 19 penalised students to 245 in two academic years, a rise of 1,190%, and the size of that jump is the least interesting thing in the file. The interesting thing is what the 245 sit inside. They are 245 of the 795 proven misconduct cases the university resolved in 2024/25, which means nearly a third of every upheld case now turns on a tool that did not exist in the regulations when the students who used it first walked through the door (The Tab). Nobody hands out a rulebook with a moving page. The penalties, however, were issued anyway, and the students who received them are the ones who absorbed the cost of an institution deciding what it meant while it was already punishing people.
The figures come from two Freedom of Information responses obtained by The Herald. In 2022/23, the first academic year in which Edinburgh recorded AI misuse as its own category, 19 students were penalised. In 2023/24 it was 78. In 2024/25 it was 245, out of those 795 proven cases — a 214% rise on the previous year and a 1,190% rise on the first year anyone bothered to count (The Herald). Three hundred and forty-two students in total, over three years, with their names attached to a finding the university itself concedes may be an undercount.
That concession is not a footnote. Edinburgh told reporters that its staff did not apply the AI sub-category consistently until 2023/24, so some AI cases from the early years were logged under plain cheating or plagiarism instead. The 342 is a floor. The real number is somewhere above it and nobody, including the university, can tell you where.
Edinburgh’s Own Tally — Academic Misconduct Cases With AI Went 19, 78, 245
Set the numerators against the denominators and the shape of the problem changes. In 2022/23, AI accounted for 3.7% of Edinburgh’s proven cases: 19 out of 515. By 2024/25 it accounted for 30.8%: 245 out of 795 (The Tab). The total volume of proven misconduct barely moved. What moved is the share of it that a chatbot is now standing behind. The university’s disciplinary machinery did not get dramatically busier. It got a new subject.
Across a student body of more than 49,000, the share involved in any proven case rose from 1.6% to 1.65% — a rounding error in percentage terms and 245 additional humans in real ones (The Herald). The penalties were mark deductions, formal warnings, or both at once. In repeated or serious cases they escalate to the Code of Student Conduct, where the consequences stop being a number on a transcript and start being a question about whether you finish the degree at all. A mark penalty is a small thing until it is the two points between a 2:1 and a 2:2, which in the UK is the difference between a graduate scheme interview and an automated rejection.
The university’s own line is that the data may not capture every case involving AI because some were filed only under the main categories. That is an admission dressed as a caveat. It means the house never finished counting before it started sentencing.
Miles Briggs — 1,051 Cases Across Scotland and One Word He Keeps Using
Edinburgh is not an outlier so much as an early-warning system. Nationally, Scottish institutions reported 131 AI-linked misconduct cases in 2022/23 and 1,051 in 2023/24 — a 700% rise across a single year, gathered under FOI by the Scottish Conservative education spokesman Miles Briggs (The Scotsman). Briggs has now looked at this data twice across two years and reached for the same phrase both times: the tip of the iceberg. He is right, and not for the reason a politician usually means it. The recorded cases are the ones a human noticed. Detection is a staffing question, and the staffing is being cut.
His argument is that the Scottish Government should set out how it intends to support universities on this, which is the kind of sentence that sounds like action and functions as deferral. The universities are already making their own policy in the gap, one institution at a time, with different definitions, different penalties and different willingness to count. A student’s exposure to a career-ending finding currently depends on which postcode their campus sits in. That is not a standards regime. It is a lottery with a disciplinary hearing attached.
Abertay and Robert Gordon — 342 Students, and 63 Who Lost Every Reassessment
The most instructive numbers in the Scottish data are not the biggest ones. Abertay University created an “unacceptable AI use” category in 2023 and dealt with 351 cases in its first full year, upholding 342 — nearly a third of the entire national total from one institution, which says more about Abertay’s detection culture than about Aberdeen’s morals (The Scotsman). Stirling investigated 213 and penalised 200. Robert Gordon University handled 116 and sanctioned every one, including 63 students who received failure and the loss of all remaining reassessment opportunities.
Read that last penalty twice. Not a lowered grade. Not a warning on file. Sixty-three people who were told the door was closed, in a year when the definition of the offence was still being drafted at national level. Some of them will have bought an essay from a ghostwriting site and deserved exactly what they got. Others will have used a grammar checker, or run a paragraph through a translation tool, and discovered afterward that the institution had decided that was false authorship. The penalty is identical because the process cannot tell them apart, and the process cannot tell them apart because it was never built to.
The Rule as Written — an AI Translator Is Now False Authorship
Edinburgh’s current student guidance, revised in October 2026, is unusually specific for the sector. Presenting AI output as your own work is misconduct. Using an AI translator to convert assessed work into English before submission is misconduct, and the document states plainly that machine translation is treated as false authorship. Submitting assessed work containing unacknowledged AI text, images, audio, video, code, formulae or reasoning is misconduct. Citing AI-sourced references without reading them is misconduct. Agentic AI operating inside any university platform that requires a login is banned outright (University of Edinburgh).
It is a serious document, and it is newer than the penalties it justifies. The guidance warns students about cognitive offloading — the atrophy of the analytical muscle you came to university to build — which is the most honest sentence in the whole policy suite, because it admits the institution understands what is actually being lost. It just took four years and 342 findings to say it out loud in a place a first-year would find it.
February 2027 — the Year Nobody Can See Yet
Edinburgh will not know its own 2025/26 numbers until roughly February 2027, because misconduct data is collated annually from three colleges and twenty-five schools after the academic year closes. Two more cohorts of students will be graded, accused, penalised and graduated before the sector has a clean picture of the year it is living through. The rulebook will be revised again in that window. It already has been, more than once. Each revision will be better than the last, and each one will land on students who were assessed under the version before it.
Axel’s Take — the Penalty Arrived Before the Definition, Every Time
Here is the part the sector will not say plainly. Universities spent four years treating AI use as a character problem when it was a specification problem. They had no shared definition, no reliable detection, no consistent penalty scale and no obligation to tell a student in advance where the line fell — and then they issued 342 findings at one institution and 1,051 across a country, and called the resulting spreadsheet evidence of a cheating epidemic. Some of it is cheating. A meaningful fraction of it is nineteen-year-olds guessing at rules their own lecturers could not have stated consistently, because the sub-category was not applied consistently until 2023/24 by the university’s own account.
The students carrying these findings are not abstractions. One of them is a third-year who used a translation tool on a draft because English is their fourth language and the writing centre was booked out, and who now has a formal warning in a file that follows them into a postgraduate application. Another read the revised guidance in October 2026, recognised something they had already submitted, and has been waiting since. Neither of them appears in a 1,190% statistic. Both of them are what it is made of. The institutions will get their definitions right eventually, the way they always do — after the people who needed them have already graduated, or not.
Sources: The Herald, The Tab, The Scotsman.