GCSE and A Level AI Malpractice Statistics

What People Are Actually Trying to Find

Searches for GCSE and A Level AI malpractice statistics generally come from journalists and researchers who want the official Ofqual numbers rather than estimates, exam centre staff and Heads of Centre preparing malpractice policy documents, parents trying to understand how common AI-related penalties actually are, and policy analysts comparing UK secondary-level enforcement with university-level AI detection policy in other countries. This page is built around the official government data first, with independent research included only where it is clearly labelled as such.

One clarification matters immediately: Ofqual’s malpractice statistics count proven cases where a penalty was issued, not suspected cases, not informal warnings that didn’t lead to a formal sanction, and not the much larger pool of AI use that may go entirely undetected. Every figure in this piece should be read with that scope in mind.

Data Sources and Methodology

Key Findings From the Official Data

The clearest finding is directional, not absolute: AI-related plagiarism roughly doubled its share of the (still small) plagiarism category in a single year, from just over half of plagiarism cases to three-quarters. That trajectory matters more than the raw case count, since it shows AI is quickly becoming the dominant form of plagiarism at GCSE and A level, even while plagiarism as a whole remains a minor share of total malpractice, behind mobile phones, disruptive behaviour, and unauthorised materials.

Total plagiarism cases (AI and non-AI combined) were 2.5% of all student malpractice in summer 2025 versus 3.0% in summer 2024, and 1.8% in summer 2023, so the category itself has not grown dramatically. What has changed sharply is its internal composition: AI misuse has gone from being roughly half of plagiarism cases to three-quarters of them in the space of one exam series.

For broader context on how these UK secondary-level figures compare with international patterns, see our related coverage of AI academic misconduct statistics and AI-generated research paper statistics for 2026.

Charts and Graphs

Full Year-by-Year Data Table

All figures below are drawn directly from Ofqual’s official statistics releases and are subject to Ofqual’s standard rounding policy.

Source: Ofqual, “Malpractice in GCSE, AS and A level” official statistics, summer 2022 through summer 2025 exam series releases, GOV.UK. AI-related plagiarism was first reported as a distinct sub-category in the summer 2024 release; prior years grouped all plagiarism together regardless of AI involvement.

Where AI Fits Among All Malpractice Types

It’s easy for a headline like “AI plagiarism up 75%” to imply AI cheating is now the leading integrity problem in English exam halls. The official data says otherwise. Mobile phones and other communication devices remain the single largest category of student malpractice by a wide margin, involved in roughly 42 to 44% of all cases in the most recent two exam series. “Other unauthorised materials,” “other reasons,” and disruptive behaviour each individually outweigh AI-related plagiarism in raw case volume. AI misuse is a fast-growing but still comparatively small slice of a malpractice landscape that is still dominated by low-tech offences.

This matters for policy discussions happening alongside detection-tool adoption, covered in depth in our companion piece on how much universities spend on AI detection tools and our list of universities that have banned AI detectors outright. Secondary-level exam boards in England have so far taken a different path from many universities: rather than banning detection software, JCQ guidance explicitly endorses its use as one tool among several for identifying suspected AI misuse.

Subjects and Assessment Types Most at Risk

Ofqual’s headline statistics don’t break AI-related plagiarism down by subject, but JCQ’s published guidance and real anonymised case examples make the risk pattern clear: AI misuse concentrates almost entirely in non-examined assessment (NEA), coursework, and word-processed exam scripts, rather than traditional closed-book, handwritten exams. That’s because NEAs give students unsupervised time to draft and redraft work outside the exam hall, exactly the conditions in which a generative AI tool can be used undetected.

Beyond these documented examples, JCQ guidance flags subjects with a heavy coursework or extended-writing component, including English, History, Religious Studies, Art and Design, and written vocational and technical qualifications, as structurally higher-risk simply because of how much unsupervised drafting time they involve. Analysts have also noted that private schools disproportionately offer qualifications like the English IGCSE, which can carry a much larger non-examined assessment weighting than the mainstream state-school GCSE, a structural difference that may concentrate AI-related risk unevenly across the school system.

Staff, Centre-Level, and Detection Software Data

AI-related integrity risk in the official statistics isn’t confined to students. Staff and whole-centre malpractice cases of all types rose sharply in the most recent exam series, driven mainly by maladministration (procedural failures without intent to deceive) rather than deliberate misconduct, but the scale of the increase is notable in its own right.

JCQ guidance requires Heads of Centre to have arrangements in place ensuring student work is authenticated properly, and states that staff who have doubts about the authenticity of a student’s work but fail to investigate can themselves face sanctions. Exam boards report using “a combination of different approaches, including detection software” to identify suspected AI use, meaning tools similar to those covered in our analysis of AI detector bias against ESL writers and AI detection false positive rates are already embedded in how UK secondary qualifications are policed, with all the reliability caveats that implies.

What the Official Numbers Don’t Capture

Every figure above counts only proven cases that resulted in a formal penalty. Independent survey research, while methodologically different from Ofqual’s administrative data and not directly comparable to it, consistently suggests the real rate of AI use in assessed schoolwork is far higher than what gets caught and penalized. Broader UK research tracking AI-related misconduct at university level found the rate of formally confirmed cheating cases rose from roughly 1.6 to about 7.5 per 1,000 students between the 2022-23 and 2024-25 periods, according to Freedom of Information data compiled by newspaper investigations, a pattern our AI academic misconduct statistics page explores in more depth.

This is also where AI detection reliability becomes a genuine fairness issue rather than a purely technical one. Our related reporting on AI detection lawsuits documents what happens when detection software gets it wrong in an assessment context, and ChatGPT hallucination statistics are a useful reminder that the same generative models producing the disputed text are themselves unreliable narrators, which complicates both the misuse case and the detection case simultaneously.

Guidance for Students and Parents

1. Understand that NEAs and coursework carry the highest documented risk. Both real JCQ case examples involved word-processed or non-examined assessment work, not supervised handwritten exams.
2. Reference any AI-assisted elements explicitly if permitted. JCQ guidance is clear that unreferenced AI content cannot be rewarded, and a false declaration of authenticity is itself treated as a separate, serious malpractice offence.
3. Keep drafts and version history. As with the university-level disputes covered in our AI detection lawsuits tracker, a documented drafting process is consistently the strongest evidence when a piece of work is challenged.
4. Know that AI detection software can misfire. Formal, structured, or advanced-vocabulary writing, exactly the register schools try to teach, can resemble the statistical signature AI detectors flag; see our detailed breakdown of AI detection false positives.
5. Ask your centre what their AI policy actually permits for research, brainstorming, and grammar support versus final-draft generation, since permitted use varies by subject and exam board.

Guidance for Teachers and Exam Centres

Why This Data Matters Beyond the Exam Hall

These figures sit inside a much larger shift in how assessment integrity is managed across every stage of education, from GCSE coursework through to AI-generated research papers at the postgraduate level. The policy responses differ sharply by country and institution type; our study of AI detection policy at 50 leading U.S. universities shows many institutions moving away from centralized detector-based enforcement, while UK exam boards have so far leaned into detection software as a standard tool. There’s also a resourcing dimension: running detection infrastructure and investigating flagged cases carries real institutional cost, examined in how much universities spend on AI detection tools, and a physical-world footprint behind the generative models themselves, covered in AI data centers and the environment.

Frequently Asked Questions

Research Sources and Further Reading

This page is built primarily on official Ofqual statistical releases and JCQ guidance documents, with independent survey research clearly distinguished from official administrative data throughout.