Plagiarism and AI at Leeds: red, amber, green — and the saved outputs you must keep

The short answer

At Leeds, every assessment carries an AI category — red, amber or green — and in amber and green assessments permitted use is only safe if you can both acknowledge it and evidence it with saved outputs from the tool. So the two things you do before you write are: find out which colour your assessment is, and start keeping your prompts and outputs.

Leeds also treats misconduct as a strict liability offence, which means the question at the finding stage is not whether you meant to break a rule. Two documents govern this: the Academic Integrity and Assistance Policy, version 1.0, approved by Senate on 8 July 2026 for implementation from 2026/27; and the Taught Students Academic Misconduct Procedure, version 5, approved by the Taught Student Education Board on 8 October 2025.

The two documents that bind you

Leeds is unusually careful about telling you what is in force, which is worth knowing because it means you can check for yourself rather than relying on a summary — including this one.

The policy carries its own status line:

“This policy was approved by Senate on 8 July 2026 for implementation from the academic year of 2026/27. This is version 1.0 of this policy which is owned by the Quality Assurance and Enhancement Team.”

Senate approval matters. It is the university’s senior academic body, so this is not a departmental guidance note that a school can quietly differ from.

The procedure carries a full version-control block:

“Title: Taught Students Academic Misconduct Procedure · Document owner: Governance and Compliance · Version: 5 · Approved by: Taught Student Education Board · Approval date: 8 October 2025”

On the face of the document: “This procedure applies to all assessments set from September 2025.” That sentence is the one to check if your case concerns older work — Leeds publishes its previous versions (2024-25, 2023-24, and the pre-2023 procedure with its penalties annex) from the same page, so the rules that applied when the assessment was set are available to you.

The two documents do different jobs. The policy says what integrity is and what assistance is acceptable. The procedure defines the offences, the penalties and the process. If you are facing a case, the procedure is the document to read line by line.

Note also that postgraduate researchers are covered by a separate document — the Postgraduate Researcher Academic Misconduct Procedure — not by the taught-student procedure described here. If you are a PGR, the tariff below is not yours.

Red, amber, green

This is the practical heart of the Leeds regime, and the procedure states it plainly at §10.19:

“The University publishes guidance on the appropriate use of Gen AI in assessments … and has a three-tier categorisation (red, amber, green).”

The procedure then gives worked examples of what counts as misuse in each tier (§10.20), and these are worth reading closely because the amber and green wordings are almost identical:

“Examples of misuse include:
a. In RED assessments: using Gen AI to create or modify work for submission.
b. In AMBER assessments: using Gen AI to create or modify work for submission in a way that is not appropriate to the assessment, that has not correctly been acknowledged or referenced, that cannot be evidenced through saved outputs from the Gen AI tool, or that breaches any stipulation on data protection and privacy in the use of GenAI.
c. In GREEN assessments: using Gen AI to create or modify work for submission that has not correctly been acknowledged or referenced, that cannot be evidenced through saved outputs from the Gen AI tool, or that breaches any stipulation on data protection and privacy in the use of GenAI.
d. In all assessments: using translation tools to prepare or modify work for submission in a way that is not appropriate to the assessment, as specified in the assessment brief.”

Four things in that passage change how you should work.

Red is absolute. There is no qualifier at all — no “in a way that is not appropriate”, no “without acknowledgement”. Creating or modifying work with Gen AI is the offence. Modifying is the word that catches people: running a paragraph you wrote yourself through a tool to tighten it is modification.

Green is not a free pass. Read (c) again. Even in the permissive tier, unacknowledged use is misuse, and unevidenceable use is misuse. Green means Gen AI is allowed, not that it is invisible.

Amber adds one thing to green: appropriateness to the assessment. In amber, the assessment brief defines the permitted scope, and going beyond it is misuse even if you acknowledge it perfectly.

Translation is caught in every tier. Line (d) applies “in all assessments”, and it points at the assessment brief for the detail. If you are writing in a second language, this is the line to check first — and it is not unique to Leeds. If you write in English as an additional language, our note on AI detectors and non-native speakers explains why this group tends to be scrutinised more, and why keeping your working record matters more for you than for anyone else.

The colours themselves live on Leeds’ own generative-AI guidance site, which the procedure cites as the authority. The categorisation of your assessment, though, comes from the assessment brief. If the brief does not say, ask the module leader and keep the reply.

Saved outputs: the evidence that decides your case

Leeds has done something quietly significant here. Rather than resting a case on a score, it has made the student’s own record the evidence — and it has published that as an express factor either way.

Aggravating factor, §15(e):

“the student cannot produce on request evidence of drafts, Gen AI outputs, evidence of proofreading annotations or other information to support their authorship of the work.”

Mitigating factor, §16(e), the mirror image:

“the student can produce on request evidence of drafts, Gen AI outputs, evidence of proofreading annotations or other information to support their authorship of the work.”

And the principle behind both, from §2 of the policy’s principles:

“Students may be asked to demonstrate authorship to confirm the authenticity of their work.”

Read those three together and the practical instruction is unambiguous. The material that protects you is material only you can generate, and only while you are working. Nobody can reconstruct a draft history after submission.

What that means concretely:

• Write in something that keeps version history — a cloud document with revisions on, or a folder of dated files, or a repository. The point is a timeline, not a technology.
• If you use a Gen AI tool in an amber or green assessment, save the outputs. Leeds names them specifically. A chat you deleted is a chat you cannot produce.
• Keep proofreading annotations — the tracked changes, the marked-up PDF, the email from the writing centre. Leeds names these too.
• Keep your reading notes and your source PDFs. They evidence the argument, not just the text.

We have written up what a convincing authorship record actually looks like in how to prove you wrote it yourself. At Leeds that guide is not general advice — it maps directly onto a published mitigating factor.

Strict liability: intention is not the question

This clause surprises people, so it is worth quoting in full (§12–13):

“Academic misconduct offences are treated by the University as ‘strict liability’ offences. This means that the University does not consider whether a student intended to commit an offence, or whether they were aware their actions were in breach of University rules, in determining whether an offence has occurred.”

“In determining whether an offence has occurred, in the first instance the University will not consider whether the student intended to gain an unfair advantage.”

Note the two carefully placed limits. Strict liability applies to whether an offence has occurred, and “in the first instance”. It does not say your circumstances are irrelevant to what happens next — Leeds publishes mitigating factors in full at §16, and “the offence is a first offence” is expressly one of them.

So the honest framing is this: “I did not know” and “I did not mean to” are not defences to the finding, but they are relevant to the penalty. If you are answering an allegation, do not build your whole response around intention. Build it around evidence of authorship — and then raise intention and circumstances as mitigation. Our guide to what to do when you are accused of using AI follows that order for exactly this reason.

If a detector score is what started this, our guide to an AI detector flagging your thesis sets out what that number does and does not establish on its own.

One more clause with a long reach, §11.4:

“Where an academic misconduct offence is suspected, the University reserves the right to investigate and take action in respect of work previously submitted by the student, even if that work has already been assessed and the marks published. This also applies after a student has graduated.”

Marks being published does not close the matter, and neither does graduating. On the other side, §7 limits the scope usefully: “The Procedure applies to all summative assessments.” Formative work — practice essays, unassessed drafts, the things that do not count towards your mark — is outside it.

What Leeds does permit — and the paid-services line

Leeds is not prohibitionist, and the policy’s section on acceptable use of assistance is refreshingly direct:

“Students are permitted to use technologies to prepare work for assessment if restrictions are not specified within the assessment brief. Technologies may be used in a supportive capacity but will not replace a student’s own learning, understanding and skills. Students should be aware that the accuracy, reliability and security of technologies cannot be guaranteed.”

“Students may use peer assistance, informal assistance and technologies for support with proofreading.”

“Students are not permitted to use paid services to prepare work for assessment.”

That last line looks like it might catch any paid tool, so read the definitions the policy attaches to it:

“Technologies are defined for the purposes of this policy as digital tools, including generative artificial intelligence (GenAI) and translation aids, that are paid, unpaid or provided by the University.”

“Paid services are defined for the purposes of this policy as an external person and/or company paid for by the student.”

So the Leeds boundary is about who does the work, not about whether money changed hands. A paid piece of software you drive yourself is a “technology”, permitted by default unless the brief restricts it. A person or a company you pay to do the work is a “paid service”, and that is prohibited outright.

Two cautions before you rely on that. First, “permitted by default” depends entirely on the assessment brief being silent — and in a red assessment it is not silent. Second, being a permitted category does not release you from the acknowledgement and saved-output requirements in §10.20.

And note where the prohibition on paid services meets the offence list: contract cheating at §10.16 expressly includes “subscription use of online tutors and online homework sites to create or modify assessments, for payment or otherwise”. A subscription is not the problem; a subscription to something that creates or modifies your assessment is.

Leeds tells you not to use detector websites

We are going to quote this even though it works against us, because you need to know it. From the Academic Integrity and Assistance Policy, the Senate-approved document, in the section on support and resources (§1.3):

“The Turnitin tutorial explains how Turnitin is used in the assessment process, and why students should not use AI detector or plagiarism detector websites.”

That is a university-wide instruction in a Senate-approved policy, and it covers third-party checkers generally — ours included. If you study at Leeds, take it at face value.

Leeds does not publish the reasonsin that sentence — they sit inside the Turnitin tutorial, which is where the policy points you. So do the tutorial. It is the university’s own explanation, and it is the one that governs your position, not ours.

The general concern behind advice of this kind, wherever you study, is worth knowing anyway: some checking services keep what you upload and compare later submissions against it, which means your own draft can come back as a match against you. Before you put work into any service, find out whether it retains it. For what it is worth, our check does not add your text to any third-party comparison database — but at Leeds, the university’s instruction is the one that governs, not our reassurance.

There is a second reason not to lean on any checker, ours included, and it has nothing to do with policy. No detector produces proof. Our own published error rates are measured on English and German corpora, and no detector — ours included — produces proof of who wrote a document. If you want the fuller version of that argument, see how accurate AI detectors really are and, for the score most students actually encounter, what a Turnitin score actually means.

The eleven named offences

Leeds defines academic misconduct broadly first — “any attempt by a student to gain an unfair advantage in assessment” — and then enumerates the offences at §10. The full list:

• Cheating in examinations (§10.2)
• Plagiarism(§10.6) — “presenting someone else’s work, in whole or in part, as their own. Work means any intellectual output, and typically includes but is not limited to text, data, images, sound or performance.”
• Re-use of own work (§10.8)
• Theft of work (§10.10)
• Collusion (§10.12)
• Fabrication of assessment or supporting evidence (§10.14)
• Contract cheating (§10.16) — “submitting work that has been obtained from third parties, whether they are inside or outside the University and whether on a commercial or non-commercial basis”, including essay mills, commissioning individuals, and the subscription services quoted above.
• Misuse of Generative Artificial Intelligence (Gen AI) and translation tools (§10.18) — a category in its own right: “submitting work that has been created or modified with the use of Gen AI and/or translation software in a way that is not permitted by the assessment.”
• Misrepresentation, fraud and impersonation (§10.22) — which expressly includes “submitting work where the content has been manipulated to avoid detection”.
• Breach of proofreading policy (§10.25) — “Except for specific exemptions, it is the policy of the University that third parties must not act as proof-readers of any academic work submitted for assessment.”
• Facilitating academic misconduct (§10.28)

Three of those deserve a sentence of explanation.

Contract cheating does not require payment. “Whether on a commercial or non-commercial basis” catches a friend, a partner or a relative producing work you submit. Nothing has to change hands.

Manipulating content to avoid detection is its own offence. It sits under misrepresentation and fraud, and it is aimed at the whole category of tools sold on the promise of getting text past a checker. At Leeds, using one is not a grey area about whether the underlying writing was yours — the manipulation itself is the offence.

The proofreading rule is stricter than most students expect. Third parties must not act as proof-readers, subject to specific exemptions. Note how that sits beside the policy line permitting “peer assistance, informal assistance and technologies for support with proofreading” — the two are reconciled by the exemptions, which the proofreading policy itself defines. If someone is going to read your thesis for you, check that document before they start, not after.

The published penalty tariff

Leeds publishes four penalty figures in the procedure (pp. 15–18), which is more than most UK universities put on the open web. Penalties scale with the level of study and with whether the case goes beyond the school.

School penalties at level 0 and level 4 — foundation and first year — are explicitly educational:

“All cases: The School will not mark the specific content in the assessment that contains academic misconduct. Students will receive a written warning and one of the following: a. The assessment will be marked with the specific content that contains academic misconduct redacted, or b. The assessment will receive a mark of zero. Students must also: Attend a meeting in the School to discuss academic skills. Engage with Skills@Library resources (website, workshop or 1-1 appointment) … Where students resit, in all cases the assessment component containing academic misconduct is capped at the pass mark.”

School penalties at all levels counting towards classification are harder, and the split turns on mitigation:

“Assessment-based penalties — In cases where there are substantial mitigating factors. … Module-based penalties — In cases where there are no substantial mitigating factors. The assessment is failed. Students are required to resit the assessment to expunge the offence and one of the following: a. The resit assessment will be capped at the pass mark. b. The resit assessment will be capped at zero. c. The resit will be capped at the pass mark for the module. d. The resit will be capped at zero for the module.”

This is where §16 stops being abstract. “Substantial mitigating factors” is the phrase that decides whether the penalty attaches to one assessment or to the whole module — and the ability to produce your drafts and saved Gen AI outputs is one of the listed factors. That is a direct line from a habit you can adopt today to a materially different outcome.

Remedial requirements attach at this level too: “Retake the Academic Integrity Test at the appropriate level … Retake Using GenAI at Leeds Training, where appropriate and specified by the School.”

Committee on Applications penalties — the stage above the school — add the serious end of the scale:

“Level and Award-based penalties. Any combination of penalties including the above and the following: a. Requirement to repeat credits or year(s). b. Reduction of classification/removal of credits. c. Exclusion with intermediate concluding qualification. d. Exclusion without award. e. Any other suitable penalty proportionate to the circumstances.”

“Exclusion without award” is the top of the tariff: removed from the university with nothing to show for the time. It sits at committee level, not school level, which is the structural protection — the most serious penalties are not in the hands of the department that raised the allegation.

The process, the standard of proof, and the OIA

Leeds publishes the whole ladder:

• Stage 1: Local Investigation, at school level.
• Stage 1a: Review of School Level Penalties.
• Stage 2: Committee on Applications — where the level-and-award penalties become available.
• Stage 3: Pro-Vice-Chancellor Review.
• Final Stage: the OIA.

The standard of proof is published, at both §29 and §48, in the same words:

“the School will determine whether there is sufficient evidence to prove the allegation(s) made on the balance of probabilities, i.e. that what is alleged is more likely than not to have happened.”

That is a lower bar than a criminal court, and it is worth being realistic about: “more likely than not” is what your evidence has to displace. It is also why a documented working record does so much work — it is not there to prove innocence beyond doubt, it is there to make the alternative account less likely than yours.

The tone of the process is published too, at §9: “All University academic misconduct case investigations are investigative rather than adversarial.” That is a real thing to hold on to when you are sitting in the meeting. You are not cross-examining anyone and nobody should be cross-examining you; the purpose is to establish what happened.

The external stage is set out in full at §57:

“If a student is dissatisfied, they may complain to the Office of the Independent Adjudicator for HE (OIA). The OIA will require evidence that a student has exhausted the University’s procedures, and that the University has issued a Completion of Procedures letter. To make a complaint a student must send the Completion of Procedures letter to the OIA within 12 monthsof the date of the letter, together with a completed copy of its complaint form …”

Two operational points from that. You cannot go to the OIA early — you must exhaust the internal stages first. And the Completion of Procedures letter is the document that starts a twelve-month clock, so put its date in your calendar the day it arrives.

Leeds publishes the review request forms themselves — a penalty appeal form and a DVC appeal submission form — from the same academic misconduct page. The procedure is administered by the Student Cases Team in the Directorate of Governance and Compliance, owned by the Head of Student Cases as nominee of the University Secretary (§85). If you need to ask who is handling your case, that is the office.

What Leeds does not publish

Leeds documents more than most, but not everything. On the public web we could not establish:

• Whether Turnitin’s AI-writing indicator is switched on or off. Leeds names Turnitin, but publishes no position either way. This is a loud silence: the Senate policy is dated July 2026 and does not mention AI detection at all.
• Whether submissions are added to a comparison repository, how long they are retained, or whether any opt-out exists. No retention statement was found in either document.
• Whether the Turnitin tutorial includes a live pre-submission similarity check for students, or is only explanatory.
• The wording of the Declaration of Academic Integrity that you make at submission. The declaration exists and is published; we have not reproduced its text here.
• Any annual case counts.

Leeds does run a mandatory AI course alongside its academic integrity tutorials and tests — the policy describes it as supporting students “with developing their skills in using GenAI, including understanding both the opportunities and the ethical challenges of the technology”. If you have not done it, that is the first place the university itself will point you.

Before you submit

• Find the colour. Red, amber or green is a property of your assessment, and it is in the brief. If it is not, ask the module leader in writing and keep the answer.
• Save every Gen AI output you used. Leeds names “saved outputs from the Gen AI tool” as the thing you must be able to produce. Unevidenceable use is misuse even in a green assessment.
• Keep drafts with dates. Being able to produce them is a published mitigating factor, and it can be the difference between an assessment-based and a module-based penalty.
• Acknowledge any permitted use. Leeds requires you to “acknowledge use of assistance”; unacknowledged use is misuse in both amber and green. Our AI disclosure statement guide shows the shape, but check your brief for a required form of words.
• Check translation separately. The translation rule applies in all assessments and points at the brief.
• Do not use an outside AI-detector or plagiarism website. Leeds says so in the Senate-approved policy. Do the Turnitin tutorial instead.
• Do not have a third party proofread itunless you have checked the proofreading policy’s exemptions first.

Sources

• University of Leeds: Academic Integrity and Assistance Policy — version 1.0, approved by Senate 8 July 2026, owned by the Quality Assurance and Enhancement Team.
• University of Leeds: Taught Students Academic Misconduct Procedure (PDF) — version 5, approved by the Taught Student Education Board, 8 October 2025; applies to all assessments set from September 2025.
• University of Leeds: Academic misconduct — Student Cases — the landing page, the appeal forms, and the previous versions of the procedure.
• University of Leeds: AI and assessments — the red/amber/green guidance, cited by the Procedure at §10.19.
• University of Leeds: Academic integrity (student hub) — the student-authored definition and the tutorials.
• University of Leeds: Postgraduate Researcher Academic Misconduct Procedure — the separate route for PGR students.

If you want to see how another UK institution has answered the same questions differently, our page on plagiarism and AI at Sheffield covers a university that has published an explicit decision not to use detection tools at all. And if you want to see what our own check does with a document — with the caveat above about Leeds’ instruction — it is at the AI check.

This page is orientation, not legal advice. What binds you is your assessment brief, the Academic Integrity and Assistance Policy approved by Senate on 8 July 2026, and the Taught Students Academic Misconduct Procedure version 5.