- What actually happens if I get caught?
- How many students are being penalised?
- What does a failed module actually cost?
Expulsion is rare and it is the wrong thing to picture. What usually happens is a zero on the assignment or a failed module, settled quietly, with no hearing and no story to tell. At the 2026/27 English fee cap of £9,790 across six modules, a failed module costs roughly £1,600 before rent and before loan interest. If a machine wrote it and nobody noticed, nothing was got away with. Something was paid for and not received.
The answer, in one line
Usually smaller and quieter than people picture. Russell Group universities recorded 2,053 punishments in 2024-25 against roughly 700 the year before, across about 350,000 students; Bristol recorded 526 penalties in 2023-24 against 7 two years earlier; Scotland recorded 1,051 cases against 131.
Definition#
The cost question: the replacement for "will I get caught?" in decisions about AI in assessed work. It asks what the assignment was bought for and whether that was received, which is answerable whatever the detection outcome, and which does not improve when the detection risk falls.
What the penalty data actually shows#
Russell Group universities recorded 2,053 punishments for AI misuse in 2024-25, against roughly 700 the year before, across about 350,000 students. Bristol recorded 526 penalties in 2023-24, against 153 the previous year and 7 in 2021-22. Scotland recorded 1,051 cases in 2023-24 against 131 the year before, of which Abertay alone accounted for 351, with 342 upheld.
Four Russell Group members are known to have expelled students: UCL, Imperial, Glasgow and Leeds. Nobody publishes a national expulsion total.
Three caveats travel with those numbers. Seven of the twenty-four Russell Group universities do not record AI investigations at all, so 2,053 is a floor on an incomplete sample. The three sets cover different years. And a rise in recorded cases mixes a change in behaviour with a change in how hard anybody is looking.
The two true things, held together#
Thousands of students are being caught. Most people who do it are not caught at all.
Both of those are supported. The penalty figures are real and rising. And at the University of Reading, 100 submissions written entirely by AI were fed into five live undergraduate modules: 94 per cent went undetected, 97 per cent on the stricter test of a marker mentioning AI, and in 83.4 per cent of comparisons the AI work scored higher than real students, an advantage of just over half a classification boundary.
Which means that if your only reason for not doing it is that you might be caught, the evidence has just weakened your reason. That is the position this page is written from, and the argument below is about money for that reason.
£1,600 a module#
The maximum tuition fee for a standard full-time undergraduate course in England is £9,790 for 2026/27, up 2.71 per cent from £9,535, uprated on the Office for Budget Responsibility's November 2025 RPIX forecast. Six modules in a year puts a single module near £1,600, before maintenance, before rent, before interest on the loan.
That figure is the whole argument. A module handed over to a model and marked as a pass is a purchase made and not collected. The receipt arrives later, in a viva, an interview or the first job where somebody asks a follow-up question.
Fees differ across Scotland, Wales and Northern Ireland, and the number of modules varies by course, so treat £1,600 as an order of magnitude rather than your own invoice.
Six ways people come unstuck, and none of them is cheating#
Almost nobody fails because they set out to cheat. They fail through a habit rather than a decision.
They use it tired. Nothing good happens at two in the morning. The version of you that accepts the first answer is the version that has been awake for nineteen hours.
They never read the output properly. Out loud, before it goes. Half the disasters on this page would have been caught by one careful read.
They mistake a good summary for understanding. It feels like learning. It is the feeling of learning without the learning, and you find out in the exam. That is the illusion of competence, and somebody has measured it.
They let it flatten their voice. Slowly, across a term. You do not notice until a tutor says your writing has got worse and you cannot explain why.
They cannot defend it in the room. The work is fine. Then somebody asks a follow-up and there is nothing underneath.
They never wrote anything down. No drafts, no version history, no notes. Innocent, and no way to show it.
Being judged on what it looked like#
The sixth habit is the one worth dwelling on, because innocence is not self-evidencing.
At a Californian graduation in June 2025, a student held a laptop up to the big screen with ChatGPT open. The stadium cheered, somebody filmed it, and within hours there were millions of views and demands that his degree be revoked. He explained the same day: he had a final due at five, the ceremony was at three, and the assessment was explicitly open to AI tools. It made no difference. A year later there were still videos claiming his degree had been revoked and inventing a fortune he had made. None of that happened.
Nothing on that list is a policy problem, and no rule would have protected him. You are judged on what it looked like rather than on what you did, which is the practical case for keeping your drafts and your dead ends: not because you are guilty, but because one day you may need to show that you are not. That is the same argument as proving you did the work, arriving twenty years earlier than it used to.
Why the penalty figures are not a rate#
The penalty figures are freedom-of-information data compiled from press reporting across different years and different institutions, not a national statistic, and they should not be read as a rate. Nobody has published what proportion of AI misuse cases end in expulsion in the UK.
The £1,600 is arithmetic on a published fee cap rather than a costing of anybody's actual module. And the six habits are drawn from teaching rather than from a study: no research has established which failure mode is most common, or in what proportion. They are offered as a description somebody may recognise, not as a measured distribution.
Key sources
- Recorded penalties for AI misuse in UK universities, compiled review.
- Scarfe, P. et al. (2024). A real-world test of artificial intelligence infiltration of a university examinations system.
- House of Commons Library (2026). Tuition fees in England. Research Briefing CBP-10155.
Related SuperSkills research#
For students, how to be honest about using AI, when your department bans it and how to use AI at university. On detection, does AI detection work. On the record you keep, proving you did the work. On the feeling that misleads you, the illusion of competence.
About this research#
The cost question and the six habits are used by Rahim Hirji in teaching and appear in the Mastering AI deck, most recently in September 2026. No claim of first use is made for either. The penalty figures, the Reading result and the fee cap belong to the sources named above. The reading offered here, that the money framing survives a falling detection risk where the fear framing does not, is an interpretation by Rahim Hirji and is marked as an interpretation and not a finding.
Evidence review · SS-2026-201 · Graded against the published rubric
Hirji, R. (2026). What actually happens if you get caught using AI. The SuperSkills evidence base, SS-2026-201. https://thesuperskills.com/research/what-happens-if-you-get-caught-using-ai. Last reviewed 6 September 2026.
An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.
How citations and IDs work