Plagiarism and AI at Edinburgh: machine translation counts as false authorship

The short answer

If you write in a language other than English and machine-translate the result before submitting, Edinburgh treats that as false authorship — academic misconduct, not a shortcut. That single rule catches more honest students than any other line in Edinburgh’s guidance, because translating your own words does not feel like passing off someone else’s work.

Where AI use is permitted, Edinburgh requires a brief acknowledgment at the end of your work, and it publishes a worked example of what that should look like. The governing texts are Using generative AI in your studies: guidelines for students, marked “Revision March 2026” and owned by Information Services, and the academic misconduct pages published by Academic Services, whose hub page states it was last published 29 July 2024.

Machine translation is false authorship

Here is the rule, verbatim, as item 2 in Edinburgh’s list of unacceptable uses:

“Using an AI translator to convert assessed work to English before submission: English is the language of teaching and assessment at Edinburgh – machine translation is treated as false authorship and is not acceptable.”

Read the reasoning Edinburgh gives, because it explains the scope. English is the language of assessment. When you are assessed, your ability to express the argument in English is part of what is being assessed. A machine that produces the English is therefore producing part of the assessed work — hence “false authorship”.

That framing tells you where the line sits in practice. Looking up a word is not producing your assessed prose. Translating a source you are quoting is a referencing question, not an authorship one. Drafting your argument in another language and running the whole thing through a translator is squarely inside the prohibition.

Edinburgh states the umbrella principle in the same document, and it is the sentence the translation rule hangs off:

“Passing off someone – or something’s – work as your own for an assessment is academic misconduct. This could be failing to cite a source you have used in a piece of assessed work, getting someone or something else (such as an AI agent) to complete your work for you, claiming authorship of machine-generated content or presenting machine-translated work as your own.”

There is also a separate prohibition for live use:

“you must not use third-party AI-based translation apps in class.”

If you are writing a thesis in your second language, this is a real constraint on how you work, and it is worth adapting to early rather than discovering at submission. Write in English from the outset, even badly, and improve it through the routes Edinburgh permits — course guidance, the study support the university provides, and the acknowledged tools described below. A rough English draft with a version history behind it is a far better position than a polished one you cannot account for. Our note on AI detectors and non-native speakers covers the wider pressure this puts on second-language writers.

A few practical consequences follow, and they are worth planning around rather than discovering in week eleven.

• Budget more time, not more tooling. Writing directly in English is slower for most people. The rule effectively converts a translation step into a drafting step, and the honest way to absorb that is in the schedule.
• Keep the English drafts. A sequence of increasingly clean English versions is the clearest possible evidence that the English is yours. A single polished file is not.
• Use your notes in whatever language you think in. The rule is about converting assessed work to English before submission. Reading, thinking and note-taking are not the submission.
• Ask about permitted language support. Edinburgh does not ban AI outright, and grammar and spelling checking appears in its own acknowledgement example. Your course guidance decides what is permitted for your assessment — so ask there rather than reasoning from the general rule.

The acknowledgement, with Edinburgh’s own example

Most universities require you to disclose AI use and then leave you to guess at the format. Edinburgh publishes an example, which removes the guesswork:

“If you choose to use generative AI for permitted aspects of an assessment, it is important to be transparent about how you have done so. You should include a brief acknowledgment at the end of your piece of work, for example:

I used GPT 5 via ELM to check grammar and spelling throughout my assignment.

I also used Midjourney to generate the image on page 2.

Again – check the detail in your course guidance if you are unsure what is required, as there may be specific things your Course Organiser would like you to cover in your acknowledgment.”

Look at the shape of the example rather than the wording. Each sentence names three things: the tool (GPT 5 via ELM; Midjourney), what it did (checked grammar and spelling; generated an image), and where (throughout the assignment; page 2). Copy that structure and you have satisfied the requirement.

Notice also what the example does not do. It does not apologise, it does not explain, and it does not editorialise. It is a factual note, placed at the end. Our AI disclosure statement guide recommends the same register for exactly this reason: a short, specific statement reads as competence, and a long one reads as a confession.

There is a second requirement on top of the acknowledgement. Where AI-generated content actually appears in your work, Edinburgh says “you will need to reference it … an in-text citation or footnote in the body of your work, and a corresponding reference in your reference list.” So the end-note covers your process; the citation covers the content. These are two separate obligations and doing one does not discharge the other.

And the standing caveat Edinburgh attaches to everything: “This guidance is general and applies to the whole University – it is essential that you also check the detailed information provided by each of your courses.” Your course guidance can be stricter. It cannot be more permissive than the misconduct rules.

The seven unacceptable uses

This list is the operative definition of AI misconduct at Edinburgh, and the framing sentence tells you what is at stake:

“The following uses of generative and agentic AI are not acceptable and constitute misconduct: if you use them you risk investigation and penalties.
1. Presenting AI outputs as your own, original work.
2. Using an AI translator to convert assessed work to English before submission …
3. Submitting an assessment which includes elements of AI-generated text without acknowledgment.
4. Submitting an assessment which includes AI-generated images, audio or video without acknowledgment.
5. Submitting an assessment which includes AI-generated mathematical formulae or reasoning, or computer code, without acknowledgment.
6. Citing and referencing AI-found sources without reading and verifying them.
7. Using an AI agent or AI browser to complete any type of work – assessed or not – within your virtual learning environment.”

Four observations that change how you should work.

Items 3, 4 and 5 are about acknowledgement, not use. The offence in each is the missing acknowledgment. That is the whole difference between a permitted workflow and a misconduct case, and it costs two sentences at the end of the document.

Item 5 covers code and mathematics. Students in technical disciplines often assume the AI rules are about prose. At Edinburgh they explicitly are not.

Item 6 catches a mistake nobody intends. Citing a source an AI tool surfaced, without reading and verifying it, is misconduct at Edinburgh even if the source turns out to be real. Language models also produce references that do not exist at all, which is a separate and worse problem — we set out how to check a bibliography in how to verify AI citations.

Item 7 applies to unassessed work too. It is the only item that reaches beyond assessment: an AI agent or AI browser must not be used to complete “any type of work – assessed or not – within your virtual learning environment”.

Set against all that, Edinburgh’s headline position is not restrictive in tone:

“The University trusts you to act with integrity in your use of AI in your learning.”

“To make the most of your time at university, embracing the hard work of learning is a better approach than looking for AI-enabled short-cuts.”

“The University does not ban the use of AI, though its use will be restricted for much of your assessed work.”

“Restricted for much of your assessed work” is the sentence to take seriously. The default for assessment is restriction, and the specific restriction comes from your course.

Notice how those two halves fit together. The trust sentence and the restriction sentence are not in tension: Edinburgh is saying it will not police you preemptively, and that the responsibility for staying inside the restriction is therefore yours. A university that trusts you is a university that will hold you to what you declared.

Which is why the acknowledgement is the centre of the whole regime here. Edinburgh has not built its position around catching people. It has built it around a declaration you make, and around what happens when the declaration does not match the work. That makes writing the acknowledgement carefully — naming everything, understating nothing — the single most protective thing you can do.

The offence categories — and where AI sits

Edinburgh publishes six definitions on its “what is academic misconduct” page:

• Plagiarism— “The presentation of another person’s work as the student’s own, without proper acknowledgement of the source, with or without the creator’s permission, intentionally or unintentionally.”
• Collusion — “a form of plagiarism. It is an unauthorised and unattributed collaboration of students in a piece of assessed work.”
• Falsification — “An attempt to present fictitious or distorted data, evidence, references, citations, or experimental results, and/or to knowingly make use of such material.”
• Cheating— “Any attempt to obtain or to give assistance in an examination or an assessment without due acknowledgement. This includes submitting work which is not one’s own.”
• Deceit— “Dishonesty in order to achieve advantage. For example, by resubmitting one’s own previously assessed work.”
• Personation — “The assumption of the identity of another person with intent to deceive or gain unfair advantage.”

Two things follow from this list. First, “intentionally or unintentionally” sits inside the plagiarism definition — an accident is still plagiarism at Edinburgh, which is why referencing carefully matters more than meaning well.

Second — and this is a structural point worth understanding — unauthorised AI use is not a separate named category at Edinburgh. Those six are the operative list. Generative AI is handled through the acknowledgement requirement in separate guidance, and an AI case is prosecuted as plagiarism, cheating or falsification depending on what happened. That is a different design from universities that have created an AI offence heading of their own — Leeds, for instance, has “Misuse of Generative Artificial Intelligence (Gen AI) and translation tools” as a category in its own right (see our page on plagiarism and AI at Leeds).

Contract cheating appears in Edinburgh’s guidance as a named term: ghostwriting services and essay mills are treated as plagiarism, “termed ‘Contract Cheating’”.

The sanction sentence on the misconduct pages is short and it is the only one published there: “All forms of academic misconduct are regarded as an offence and are punishable under the University’s Code of Student Conduct.” That names the instrument, not the penalty — more on that gap below.

Proofreading tools

Grammar and language tools are where most students actually meet this policy, and Edinburgh handles them in two places.

The misconduct page says that proofreading tools which use AI “should be appropriately acknowledged, as when using other generative AI tools”. So they are not outside the acknowledgement requirement — Edinburgh’s own worked example is, after all, about checking grammar and spelling.

The guidance then points to a dedicated document for the detail:

“For more detailed information about the restrictions around using AI-supported online proofing tools, please see the University’s Guidance on Proofreading of Student Assessments (pdf).”

Note the word “restrictions”. Edinburgh is signalling that there are limits, not merely a disclosure duty, and it puts them in a separate document. If you use a proofreading tool on assessed work, read that PDF before you rely on this page or any other summary. It is linked from the misconduct pages and is publicly available.

ELM, and why Edinburgh steers you to it

Edinburgh runs its own hosted LLM service, ELM, and its reasoning for pointing students at it is about data and access rather than detection:

“In ELM your data is secure – it will not be retained by third party services to train their models or for any other purpose.”

“It is free to use for all staff and students, providing the same access for all and saving you money.”

Both points are worth taking seriously beyond Edinburgh. Unpublished thesis material handed to a consumer service may be retained; that is a genuine risk to your own unpublished work, quite apart from any integrity rule. And the equal-access argument is the university acknowledging that a paid tier creates an unfair advantage.

Using ELM does not remove the acknowledgement duty. Edinburgh’s own worked example says “I used GPT 5 via ELM” — the university’s own service, and still declared.

Postgraduate researchers

If you are a doctoral or research student, the guidance above is not quite yours. Edinburgh publishes Generative AI Guidelines for Postgraduate Research Students as a separate document on the same index, and it draws the line differently: the University “does not ban generative AI use in research but restricts it for assessed work”.

That distinction is worth sitting with, because a PhD blurs the two. Your research practice and your assessed submission are different activities, and a tool that is unremarkable in the first can be a problem in the second. The thesis you submit for examination is assessed work.

The PGR guidance also says that programme- and supervisor-level guidance may add restrictions on top. So there are potentially three layers you have to satisfy: the University’s guidelines, your programme’s, and your supervisor’s. The safe order is to ask your supervisor first and record what they say — a supervisor’s written answer is both the most restrictive layer and the easiest to obtain.

Nothing in what we could read suggests the acknowledgement requirement is relaxed for research students. If anything, a thesis makes it more important: it is a long document, examined by people who will ask you about it in a viva, and an acknowledgement written at the end is a great deal easier than an explanation improvised in the room.

Who decides your case

Edinburgh publishes the decision-making structure, and it is decentralised:

“School and College Academic Misconduct Officers (SAMOS/CAMOS) who are responsible for the investigation at School or College level, and for determining appropriate penalties.”

So your case is investigated and decided at school or college level, not by one central university panel, and the same officers who investigate also determine the penalty. Edinburgh names the College Academic Misconduct Officers for each of its three colleges on that page, so you can find out who has responsibility for your area.

The appeal route is stated, though indirectly:

“You have the right to appeal decisions made by a Board of Examiners, including decisions affected by the outcome of an academic misconduct investigation”

— through the University’s Student Appeal Regulations. Read that carefully, because the shape matters: what you appeal is the Board of Examiners’ decision, and the misconduct outcome comes into it as something that affected the decision. If you are appealing, get the Student Appeal Regulations and work to their grounds and their deadlines.

If you are at this stage, our guide to what to do when you are accused of using AI sets out what to gather and in what order, and how to prove you wrote it yourself covers the evidence that actually persuades a panel.

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.

What Edinburgh does not publish

We would rather tell you what is missing than guess at it. Working from the pages readable to us, we could not establish the following.

• The procedure document itself.Edinburgh’s Academic Misconduct Procedure is served from the University’s SharePoint policy repository and asks for authentication. It exists and is published — we simply could not read it, so we cannot give you its version number or approving body. The procedure page describes it as applying after 16 September 2024. If you are in a case, sign in and read it; it is the document that governs, not this page.
• Any penalty tariff.No scale of penalties was found. The only operative statement is that misconduct is “punishable under the University’s Code of Student Conduct”. Unlike some UK universities, Edinburgh does not put a published table of outcomes on the open web, so we cannot tell you what a first offence typically costs. Ask, or read the procedure once signed in.
• Which similarity tool is used. No product is named on the academic misconduct pages, and we could not locate a learning-technology page describing one.
• Whether AI-writing detection is used. No statement either way. This is a notably loud silence: the student guidelines were revised in March 2026 and list prohibited uses in detail, yet say nothing about detection software. Do not read that as “no detector”; read it as “not published”.
• The standard of proof applied.
• Whether submissions are added to a comparison repository, and any retention or opt-out.
• Whether any student self-check route exists.
• Any external stage after the internal procedures. The Office of the Independent Adjudicator is not mentioned on these pages, and we are not going to tell you which external body applies to a Scottish institution when Edinburgh does not say. Ask the university for the position in writing if you reach that point.
• Any annual case counts.

One structural note: Edinburgh’s academic services pages moved, so www.ed.ac.uk/academic-services/… now redirects to registryservices.ed.ac.uk/academic-services/…. Both are institutional. If an old link from a course handbook or a forum post fails, that is usually why.

Because no detector position is published, the question “will a checker clear me?” has no useful answer at Edinburgh. It would not have a good one anyway: our own published error rates are measured on English and German corpora, and no detector — ours included — produces proof of authorship. The longer version is in how accurate AI detectors really are. What Edinburgh has published is what it expects from you: an acknowledgement, honest authorship, and no machine translation.

Before you submit

• Do not machine-translate assessed work into English. Edinburgh treats it as false authorship. Write in English from the start.
• Add the acknowledgment at the end, in Edinburgh’s own shape: which tool, what it did, where.
• Cite AI-generated content separately — an in-text citation or footnote plus a reference-list entry. The end-note does not replace this.
• Read and verify every source before you cite it. Citing an AI-found source unverified is on the unacceptable list.
• Check your course guidance. Edinburgh says twice that the university-level guidance is general and your course sets the detail.
• Read the proofreading guidance PDF before using an AI-supported proofing tool on assessed work.
• Keep your drafts. With no published detector position and no published tariff, the record of how the work was made is the most useful thing you can hold.

Sources

• University of Edinburgh: Using generative AI in your studies: guidelines for students — marked “Revision March 2026”, owner: Information Services. The unacceptable-use list, the translation rule and the acknowledgement example.
• University of Edinburgh: Academic misconduct (hub) — page states it was last published 29 July 2024.
• University of Edinburgh: What is academic misconduct — the six category definitions and the note on AI proofreading tools.
• University of Edinburgh: Academic misconduct procedure — the SAMOS/CAMOS structure and the appeal route. The procedure document itself sits in the University’s SharePoint repository and requires sign-in.
• University of Edinburgh: Guidance on Proofreading of Student Assessments (PDF).
• University of Edinburgh: ELM — the University’s own hosted LLM service.

If you want to compare regimes, our page on plagiarism and AI at Sheffield covers a university that has published an explicit decision not to use detection tools, and plagiarism and AI at Leeds covers one with a full published penalty tariff. If you want to see what a check reports on a document of your own, ours is at the AI check.

This page is orientation, not legal advice. What binds you is your course guidance, the University’s generative AI guidelines for students, the Academic Misconduct Procedure, and the Code of Student Conduct.