Quick answer: The latest academic misconduct statistics in higher education show recorded misconduct shifting from copying to AI misuse.

A Guardian Freedom of Information investigation across UK universities found almost 7,000 proven cases of cheating with AI tools in 2023-24, or 5.1 per 1,000 students, up from 1.6 per 1,000 the year before, while proven cases of traditional plagiarism fell from 19 to 15.2 per 1,000. Self-report research suggests the real rate of serious misconduct is far higher than the recorded rate: a systematic review of contract cheating found that 15.7% of students in studies from 2014 onwards admitted paying someone to do their work, and a blind test at a UK university found that 94% of AI-written exam submissions went undetected.

Academic misconduct statistics in higher education come from three very different kinds of evidence, and confusing them is the most common mistake in essays, policy papers and thesis literature reviews on the topic. Recorded cases tell you what institutions caught and proved. Self-report surveys tell you what students admit to researchers. Detection and infiltration studies tell you how much gets through. This page sets out the best current figure from each kind, what it can and cannot support, and how to cite it.

Academic misconduct statistics in higher education at a glance

Each figure answers a different question. Read the sections below before you put any two of them in the same sentence.

Recorded cases: what UK universities proved

The most detailed recent picture of recorded misconduct comes from a Guardian investigation published in June 2025. The newspaper contacted 155 UK universities under the Freedom of Information Act asking for proven cases of academic misconduct, plagiarism and AI misconduct over five years. Of these, 131 provided some data, although not every university had records for each year or category.

The headline findings were:

• Almost 7,000 proven cases of cheating using AI tools in 2023-24, equivalent to 5.1 for every 1,000 students, up from 1.6 per 1,000 in 2022-23.
• Figures up to May 2025 suggested a further rise to about 7.5 proven AI cases per 1,000 students.
• Confirmed cases of traditional plagiarism fell from 19 per 1,000 students to 15.2 in 2023-24, and early figures pointed to about 8.5 per 1,000.
• In 2019-20, before generative AI was widely available, plagiarism accounted for nearly two-thirds of all academic misconduct.
• More than 27% of responding universities did not yet record AI misuse as a separate category of misconduct in 2023-24.

Two cautions apply. First, these are proven cases, so they reflect how hard each type of misconduct is to prove as much as how common it is. Copied text can be matched to its source, while AI-generated text usually cannot be traced to anything. Second, the categories are still being built: where a university had no AI category, AI cases may sit inside plagiarism or general misconduct, which would understate the shift.

Self-report studies: what students admit

Recorded cases are a floor. To estimate how much misconduct actually happens, researchers ask students directly, usually anonymously.

Contract cheating. The most cited synthesis is Philip M. Newton’s 2018 systematic review in Frontiers in Education, which pooled 71 samples from 65 studies dating back to 1978, covering 54,514 students. Across all samples, 3.52% of students self-reported paying someone to do their work. Samples from 2014 onwards averaged 15.7%, and the review found a significant positive correlation between year and self-reported contract cheating. Newton also flags the limits of the evidence: most samples used convenience sampling, the median response rate among samples that reported one was 17.2%, and many did not make clear to students that responses were anonymous. He concludes that contract cheating is likely under-reported.

Australia. A large national study by Bretag and colleagues, published in Studies in Higher Education in 2019, surveyed 14,086 students across eight Australian universities. About 6% (814 students, 5.78% of respondents) reported engaging in one or more of five cheating behaviours associated with outsourcing their work. The site’s deeper dive into contract cheating and essay mill statistics covers this literature in more detail.

AI text. The HEPI and Kortext Student Generative AI Survey 2026, based on 1,054 UK full-time undergraduates polled in December 2025, found that 12% had included AI-generated text directly in a piece of assessed work, up from 8% in 2025 and 3% in 2024. Whether that counts as misconduct depends on the assignment’s rules, since some courses now require AI use, but it is the clearest self-report trend line available.

Detection and infiltration studies: how much gets through

The gap between self-reported and recorded misconduct is explained largely by detection. Two kinds of study measure it.

Infiltration tests. Scarfe, Watcham, Clarke and Roesch (2024, PLOS ONE) injected entirely AI-written submissions into the examinations system for five undergraduate psychology modules at a UK university, without markers knowing. 94% of the AI submissions were not detected, and the AI answers scored on average half a grade boundary higher than real students’ work. Across modules there was an 83.4% chance that the AI submissions would outperform a random selection of the same number of real student submissions.

Classifier data. Turnitin reported in an April 2024 press release that of more than 200 million papers reviewed by its AI writing detector up to March 21, 2024, about 11% had at least 20% AI writing present and about 3% had at least 80%. These are flags from a probabilistic model, not proven cases, and the research on false positives, especially for non-native English writers, means a flag is not evidence on its own. The site’s summary of AI detection accuracy statistics covers that research.

Why recorded and real rates diverge

Putting the evidence together, recorded AI cases of around 5 per 1,000 students sit next to a self-reported direct-inclusion rate of 12% and an infiltration test in which 94% of AI submissions passed unnoticed. Several mechanisms explain the gap:

1. Proof standards. Misconduct panels need evidence. Plagiarism leaves a source to match against; AI text does not, so many suspected cases are never brought or never proven.
2. Reluctance to accuse. Because detectors produce false positives, many institutions advise staff not to act on a score alone, which is the right protection for honest students but also lowers recorded rates.
3. Category lag. More than a quarter of responding UK universities had no separate AI category in 2023-24, so AI cases are scattered across other headings.
4. Rule variation. The same behaviour, such as editing AI-generated text, may be required on one module and prohibited on another, so “misconduct” is not a single fixed behaviour.
5. Self-report bias. Surveys may undercount prohibited behaviour, particularly where anonymity is unclear, as Newton’s review notes.

The student experience of misconduct processes

The statistics are not only about cheating. HEPI’s 2026 report notes that 65% of students say assessment has changed significantly in response to AI, and that some students describe anxiety about being falsely accused of misconduct. That anxiety is rational: when proof is difficult and detection is imperfect, honest students carry some of the risk. If you are ever asked to explain your work, the strongest protection is a visible process, meaning drafts, notes, version history and a clear record of which sources you read. The guide to reducing AI detection false positives on a thesis sets out what that record should contain.

Using these statistics in a thesis or literature review

If you cite academic misconduct statistics in higher education in your own work, take care: academic integrity is a popular thesis topic in education, psychology and higher-education policy, and examiners mark down literature reviews that mix incompatible figures. A few rules keep your use of the data defensible:

• Name the type of evidence. Say whether a figure is a recorded case rate, a self-report rate or a detection result, and do not compare across types without saying so.
• Give the denominator. “5.1 per 1,000 students” and “12% of respondents” are not on the same scale; state both the rate and the population.
• Date the fieldwork. The Guardian data cover academic years up to 2023-24 with partial 2024-25 figures; HEPI 2026 was fielded in December 2025.
• Cite the primary source. Use Newton (2018), Bretag et al. (2019) and Scarfe et al. (2024) directly rather than a news summary, and cite the Guardian investigation as journalism, not as a peer-reviewed dataset.
• Report limitations. Convenience samples, low response rates and inconsistent recording categories should appear in your discussion of the evidence, not be hidden.

If you are writing the review itself, the guide to writing an AI-use declaration for your thesis is also worth reading, because integrity researchers are increasingly asked to model good practice in their own documents.

What the trend means for institutions and students

The direction of travel is clear even where the exact level is not. Traditional plagiarism is falling as a share of recorded misconduct, AI misuse is rising quickly, and the proportion of misconduct that is detected and proven appears to be shrinking. Institutions are responding by redesigning assessment rather than relying on detection, and HEPI’s 2026 report recommends a mix of assessments that exclude AI and assessments that expect students to use it. For students, that means the rules may differ between modules, so the safest habit is to check the instructions for each assessment and keep a record of how you produced your work.

For a related guide, see how many students use ChatGPT for essays.

Where Tesify fits

A transparent process is the best defence against both misconduct and false suspicion. Start structuring your thesis in Tesify. It helps you structure your thesis and keep your chapters organized as you write them, and 9,000+ students have written 15,000+ chapters with Tesify. Every word stays 100% written by you.

Frequently asked questions (FAQ)

How common is academic misconduct in higher education?

It depends on the measure. UK universities proved about 5.1 AI cheating cases per 1,000 students in 2023-24, while self-report studies find much higher rates, such as 15.7% admitting contract cheating in samples from 2014 onwards in Newton’s 2018 review.

Is AI cheating overtaking plagiarism?

In recorded UK cases the gap is closing fast. Proven AI cases rose from 1.6 to 5.1 per 1,000 students between 2022-23 and 2023-24, while proven plagiarism fell from 19 to 15.2 per 1,000, according to the Guardian’s FOI data.

Why are recorded misconduct rates so low compared with surveys?

Recorded rates count only proven cases. AI-generated text is hard to prove, detectors produce false positives, and many universities did not record AI misuse as a separate category, so many cases are never recorded.

How many AI-written submissions go undetected?

In a blind test at a UK university reported by Scarfe and colleagues in PLOS ONE in 2024, 94% of fully AI-written exam submissions were not detected by markers.

What percentage of students pay someone to write their work?

Newton’s systematic review found 3.52% across all samples since 1978 and 15.7% in samples from 2014 onwards, while Bretag and colleagues found about 6% reporting cheating behaviours in a survey of 14,086 Australian students.

Are Turnitin AI scores a measure of misconduct?

No. Turnitin’s figures show how many papers its classifier flagged, about 11% with at least 20% AI writing up to March 2024, but a flag is a probability, not proof, and honest work can be flagged.

Which sources should I cite for misconduct statistics in a thesis?

Cite peer-reviewed studies such as Newton (2018), Bretag et al. (2019) and Scarfe et al. (2024) for research findings, and cite the HEPI survey and the Guardian investigation as survey and journalism sources, giving dates and populations for each.

What should I do if I am wrongly suspected of misconduct?

Ask which evidence is being relied on, provide your drafts, notes and version history, and follow your institution’s appeal or representation process, ideally with support from your students’ union or advice service.