In summer 2025, Ofqual counted 100 cases of plagiarism involving AI among 5,025 cases of student malpractice in GCSE, AS and A level. Phones and other devices accounted for 2,225. Those are the regulator's own figures, published in December 2025.
AI is the case everyone talks about and, so far, a small part of the numbers. What has grown is the anxiety around it, and anxiety is where the mistakes happen.
The first mistake is treating a detector score as a finding. JCQ does not require detection tools. Where a centre uses one, JCQ says its output should be one part of a wider look at whether the work is the candidate's own, with all the available information weighed. Ofqual's advice note to awarding bodies, dated 27 April 2026, says the same: such tools are sources of evidence, not the thing that decides.
Young people have noticed when this goes wrong. In Public First's survey of 4,021 people aged 11 to 24 in England, carried out in August 2026, 13.2 per cent of secondary pupils and 24.9 per cent of college and sixth form students said a teacher had wrongly accused them of using AI in the past year. Nearly half of secondary pupils did not know whether their work goes through a checker at all.
So a score goes in the file beside the drafts, the classwork and the teacher's account. It is never put to a candidate on its own. The guide says why, with the passages from JCQ and Ofqual: https://schoolaigovernance.co.uk/guides/jcq-ai-coursework#detectors-are-evidence-never-the-verdict
The second mistake is thinking every suspicion must go to the exam board. It depends on one signature. Until the candidate has signed the declaration of authentication, a concern about the content of the work, AI misuse included, is handled under the centre's own procedures, with details recorded. Once the declaration is signed, the head of centre must notify the awarding body immediately, and the awarding body decides the sanction.
That switch is the most useful fact for anyone supervising coursework, because it rewards raising a doubt early. Before the signature, it is a teaching conversation: gather what exists, ask the candidate to talk the section through as a question rather than a verdict, decide under the malpractice policy and record it. After the signature, it is a formal referral on form JCQ/M1. The route is set out here: https://schoolaigovernance.co.uk/guides/jcq-ai-coursework#the-declaration-decides-the-route
Two of this year's changes deserve a line each. Teachers must now review all of a candidate's work and decide the mark themselves; AI must not be the sole or primary means of marking. And in a paper-based exam where a candidate uses a word processor, the centre must switch off every AI tool the candidate could reach, in the application, in the operating system and in any browser, and test that before the exam. The exams officer and whoever manages IT need an afternoon together before the mocks.
The order of work each September is in the guide: policy first, then staff, then candidates, before any assessment starts. Every candidate must receive JCQ's information for candidates on AI before assessments take place, and a dated record of who received it answers the M1's question before anyone asks it: https://schoolaigovernance.co.uk/guides/jcq-ai-coursework#the-september-calendar
The guide covers GCSE, AS and A level non-examination assessment and Project qualifications, with every regulation and instruction cited by paragraph and the date it was read. It is free: https://schoolaigovernance.co.uk/guides/jcq-ai-coursework
For the candidates themselves there is a one-page briefing, Candidates and AI, which says what they need to know before the work starts, in words a Year 10 can read. It is free to download and adapt: https://schoolaigovernance.co.uk/templates/candidates-and-ai
Wes Akram is an exams officer, data manager and DPO at a secondary school, and the author of The School AI Governance Handbook: for the slightly overwhelmed. Everything on schoolaigovernance.co.uk is free.
The facts behind this article
Every claim above is numbered to one of these records. Each gives the source’s own words where it can, the paragraph or section, and the date it was last read.
- 01
Ofqual counted 100 cases of plagiarism involving AI in summer 2025, 2.0 per cent of 5,025 student malpractice cases; phones and other devices accounted for 2,225.
Malpractice in GCSE, AS and A level: summer 2025 exam series, the release, tables 6.1 to 6.5. Dated 11 December 2025. Read 28 September 2026.
- 02
In summer 2025, across GCSE, AS and A level, 1,125 malpractice penalties cost a student a whole qualification (loss of aggregation or certification opportunity) and 1,955 took marks away; 4,735 students received at least one penalty. These are penalties, not students.
Malpractice in GCSE, AS and A level: summer 2025 exam series, the release, 11 December 2025. Dated 11 December 2025. Read 28 September 2026.
- 03
JCQ does not require AI detection tools, names some as examples without endorsing them, and says their use should form part of a holistic approach to the authenticity of students' work, with all available information considered. A detector score is never proof on its own; JCQ does call detectors a useful part of the evidence, so "not accepted as proof of anything" overstates it.
“The use of detection tools, where used, should form part of a holistic approach to considering the authenticity of students' work”
AI Use in Assessments: Your role in protecting the integrity of qualifications, section 8, page 13. Dated 30 April 2025. Read 28 September 2026.
- 04
Ofqual's advice note to awarding bodies of 27 April 2026 says detection tools should be used as sources of evidence rather than as sole determinants, with the risk of false positives in mind.
“should be used as sources of evidence rather than as sole determinants”
Artificial intelligence malpractice and assessment: advice note, the note. Dated 27 April 2026. Read 28 September 2026.
- 05
Public First's poll for the report was an online survey of 4,021 people aged 11 to 24 in England, carried out from 18 to 27 August 2026, weighted by age, gender and region. "Secondary" means Years 7 to 11; "college" means college and school sixth form. Unweighted bases: 1,326 secondary pupils and 930 college and sixth form students unless a table says otherwise.
AI in Education: polling tables (Public First for Seldon and Bunting), methodology sheet. Dated 22 September 2026. Read 27 September 2026.
- 06
13.2 per cent of secondary pupils and 24.9 per cent of college and sixth form students say a teacher has wrongly accused them of using AI.
AI in Education: polling tables (Public First for Seldon and Bunting), Table 87. Dated 22 September 2026. Read 27 September 2026.
- 07
21.0 per cent of secondary pupils and 36.4 per cent of college students say their work goes through an AI checker; 45.3 per cent and 26.7 per cent do not know whether it does.
AI in Education: polling tables (Public First for Seldon and Bunting), Table 93. Dated 22 September 2026. Read 27 September 2026.
- 08
The candidate's declaration of authentication confirms the work is their own, sources are referenced and any AI use is acknowledged, and is signed when the work is finished; the teacher confirms the work is solely the candidate's, done under the required conditions, and, on the 2026/27 form, that reasonable steps were taken to reduce the risk of AI misuse.
Instructions for conducting non-examination assessments (GCE and GCSE) 2026/27, Appendix 4. Dated September 2026. Read 28 September 2026.
- 09
Failing to report suspected malpractice is itself malpractice. Breaches of the assessment conditions are reported whenever they are found; concerns about the content of the work found before the candidate signs are dealt with under the centre's own procedures, with details recorded.
Instructions for conducting non-examination assessments (GCE and GCSE) 2026/27, section 9 (with jcq-suspected-malpractice-2026, 4.5). Dated September 2026. Read 28 September 2026.
- 10
Once the declaration is signed, the head of centre must notify the awarding body immediately on form JCQ/M1 of any suspected malpractice, and the awarding body decides the sanction, from zero for the piece of work up to disqualification. The M1 asks whether the AI information sheet was issued before the declaration was signed.
Suspected Malpractice: Policies and Procedures, 4.5; the M1 form at Appendix 8. Dated September 2026. Read 28 September 2026.
- 11
For 2026/27, teachers must review all of a candidate's work and determine the mark it warrants regardless of the outcome of any AI tool, and the teacher remains responsible for the mark or grade awarded. AI must not be the sole or primary means of marking.
“The teacher remains responsible for the mark/grade awarded.”
General Regulations for Approved Centres 2026/27, 5.3(b) (NEA instructions 2026/27, 6.1, says the same). Dated August 2026. Read 28 September 2026.
- 12
New for 2026/27: in a paper-based exam where a candidate uses a word processor without internet access, the centre must turn off every AI tool the candidate could reach, in the word processor, the operating system (Copilot, Apple Intelligence and the like) and any browser or app; centres must not assume that turning one of these off is enough on its own, and the set-up must be tested before the exam.
“Centres must not assume that turning one of these off is enough on its own”
Instructions for conducting examinations 2026/27, 14.25 (listed in the key-changes table). Dated September 2026. Read 28 September 2026.
- 13
Centres must give candidates the JCQ information for candidates documents, the AI one included, before assessments and examinations take place.
General Regulations for Approved Centres 2026/27, 5.8(c). Dated August 2026. Read 28 September 2026.