Plagiarism and AI at Columbia: “Unauthorized Use of Artificial Intelligence” is its own violation

The short answer

Columbia College and Columbia Engineering list “Unauthorized Use of Artificial Intelligence” as its own named violation, and the default is prohibited: AI is off the table for an assignment unless your instructor has given a clear statement granting permission.

Two things follow that most students miss. A violation “may be intentional or unintentional”, so meaning well is not a defence. And the finding is made on the balance of probabilities — the “preponderance of evidence” — through the Dean’s Discipline process, not by a tool.

Which Columbia this is

There is no single Columbia rule, so the scope has to be stated first. Everything below is the Columbia College and Columbia Engineering (SEAS) undergraduate regime, administered by Student Conduct, a unit within the Center for Student Success and Intervention (CSSI).

The Graduate School of Arts and Sciences, the School of Professional Studies and the professional schools set their own rules, and those hosts were not readable to us. There is also a university-level AI policy — the Center for Teaching and Learning links to one at the Office of the Senior Vice Provost — and that page was blocked to us as well, so we cannot tell you what it says. If you are not in CC or SEAS, treat this page as context and find your own school’s statement.

One further caveat on dating: none of the Columbia pages quoted here carries an effective or last-revised date. Their footers read © 2026, and the CTL page self-cites as 2026. Where a document dates itself, we say so.

The Columbia College Honor Code is unusual in where it comes from — it was adopted by student resolution rather than faculty vote:

“The Columbia College Student Council, on behalf of the whole student body, has resolved that maintaining academic integrity is the preserve of all members of our intellectual community – including and especially students. As a consequence, all Columbia College students make the following pledge: We, the undergraduate students of Columbia University, hereby pledge to value the integrity of our ideas and the ideas of others by honestly presenting our work, respecting authorship, and striving not simply for answers but for understanding in the pursuit of our common scholastic goals.”

The affirmation students make is shorter: “I affirm that I will not plagiarize, use unauthorized materials, or give or receive illegitimate help on assignments, papers, or examinations. I will also uphold equity and honesty in the evaluation of my work and the work of others.”

The named AI violation

Columbia CC/SEAS does not leave AI to interpretation. It sits in the violation schedule itself, alongside Plagiarism, Cheating, Fabrication and eighteen others:

“Unauthorized Use of Artificial Intelligence — Absent a clear statement from a course instructor granting permission, the use of generative Artificial Intelligence (AI) tools to complete an assignment or exam is prohibited. The unauthorized use of AI shall be treated similarly to unauthorized assistance and/or plagiarism.”

Take the sentence apart, because each clause does work.

“Absent a clear statement…”— the default is prohibition. A syllabus that says nothing about AI is not permissive. If you are unsure whether your course allows a tool, the published answer is that it does not, until an instructor says otherwise. And “clear statement” sets the bar: a vague remark in a lecture is thinner ground than a written line in the syllabus. Ask by email and keep the reply.

“…from a course instructor”— permission is per course, from the person teaching it. One instructor’s allowance tells you nothing about the next.

“to complete an assignment or exam” — the rule attaches to completing the work. It does not enumerate stages the way some universities do, and it does not carve out brainstorming or outlining. That ambiguity cuts against you, so if you plan to use a tool at an early stage, ask whether it is covered.

“shall be treated similarly to unauthorized assistance and/or plagiarism” — this tells you the consequences. It is not a lesser, novel category with its own gentler handling; it slots into the doctrine Columbia already applies to getting help you were not allowed.

The Faculty Statement on Academic Integrity, written before the AI category existed, already reached in the same direction with one phrase: “you must always submit your own work and not that of another student, scholar, or internet agent.”

If your course does permit AI use and asks you to say so, no citation format is prescribed on Columbia’s own domains that we could find. Our AI disclosure statement guide covers what a usable one contains; the form your instructor asks for wins over any template.

What else Columbia counts as dishonesty

The AI category does not stand alone. It is one entry in a violation schedule that runs to more than twenty, sitting beside Plagiarism, Cheating and Fabrication. Columbia also publishes a longer student-facing explainer that groups the same ground into plagiarism, cheating, fabrication, unauthorised sources of assistance, assisting others, and “other forms”, each with worked examples.

Two of those groups deserve your attention if AI is the thing on your mind, because they are where an AI question will actually be argued.

Unauthorised sources of assistanceis the doctrine the AI rule points at by name — “shall be treated similarly to unauthorized assistance and/or plagiarism”. So the question a dean asks is not a novel technological one. It is the old question: did you get help you were not permitted to have on work presented as your own? That framing is worth internalising, because it tells you what a good answer looks like. Not “the detector is wrong”, but “here is how this work was produced, and here is what I was permitted to use”.

Assisting othersis the one students underestimate. Columbia’s examples in this area are blunt and they run in both directions: “Pay someone to write a paper for you”, “Allow someone else to write a paper for you”, “Purchase a pre-written paper”. Note the second one. Letting a friend write for you is listed on its own, which means the person who helps is exposed as well as the person who submits. If a classmate asks you to run their draft through a rewriting tool, that request is not harmless to you.

Those examples are also a useful measure of where the line has historically sat. Columbia’s picture of dishonesty is someone else’s writing arriving in your submission. A generative tool producing prose you then hand in fits that picture exactly, which is why Columbia was able to assimilate AI into existing doctrine rather than invent a new one for it.

And it tells you what is not in the picture. Nothing in these categories reaches the ordinary business of writing: reading, taking notes, drafting, revising, asking a writing centre for feedback, running a spellchecker. If you did the thinking and the sentences are yours, you are inside the rules as Columbia has written them, whatever any tool later reports about your prose.

Intent is not required

This deserves its own section because it changes how you should prepare:

“Violations of academic integrity may be intentional or unintentional and may include dishonesty in academic assignments or in dealing with University officials, including faculty and staff members. Moreover, dishonesty during the Dean’s Discipline hearing process may result in more serious consequences.”

“I did not mean to” is not a defence to the finding. It may well matter to the sanction — the process weighs “the specifics of the case… aggravating circumstances” — but it does not answer the question of whether a violation occurred. A citation you genuinely forgot, a paraphrase you thought was far enough from the source, a tool you did not think counted: all of these can be violations on Columbia’s own terms.

The practical response is not fear, it is care at the point of writing. Track where every sentence came from. Keep the drafts. If you used any tool at all, write down which and for what, before anyone asks. See how to prove you wrote it yourself.

The second sentence is the one to take most seriously if a meeting is ever scheduled. Dishonesty during the process is treated as its own aggravating matter. An account that has to be corrected later is worse than a difficult truth told at the start.

When your syllabus does not say

Columbia puts the obligation to define the AI rule on the instructor, not on the university. Its Center for Teaching and Learning gives faculty a pre-publication checklist for their syllabus, and two of the boxes are yours to care about:

“[ ] Have I addressed whether generative AI is allowed, limited, or prohibited?
[ ] Have I explained what students must disclose?”

So Columbia expects your syllabus to answer two questions: allowed, limited or prohibited — and what you have to say about it if you use anything. Note the middle option. “Limited” is a real category, and it is the one that produces most of the confusion: a course may permit AI for, say, generating practice questions but not for drafting prose. If your syllabus says “limited” without saying limited to what, that is a question, not an invitation to guess.

Plenty of syllabi will not have both boxes ticked. When yours does not, the default in the violation schedule takes over and the answer is no — “absent a clear statement… prohibited”. There is no Columbia-wide permission underneath the silence for you to fall back on.

The practical move is a short, unembarrassing email, sent before you start rather than after you submit. Something like: my syllabus does not mention generative AI — may I use it for X, and if so, how would you like me to disclose it?Three things happen. You get an answer. The answer is in writing, which is what “a clear statement from a course instructor” means. And if a question is ever raised later, you can produce the exchange.

Be specific about the X. “May I use AI?” invites a cautious no. “May I use a tool to check my grammar on a draft I wrote myself?” asks something a teacher can actually answer, and the answer is more useful to you.

Keep the reply for as long as you might need it. An instructor moves on, a course page comes down at the end of term, and a permission you cannot produce is a permission you did not get. A screenshot in the folder with the assignment costs you nothing.

The one tool Columbia names

Columbia’s Center for Teaching and Learning lists, under “Resources → Columbia Integrity Policies”:

“CUIT: Turnitin Services”

So Turnitin is named on Columbia’s own domain, as a service run by Columbia University Information Technology. That is more than most universities publish.

It is also all Columbia publishes that we could read. The CUIT service page itself returned HTTP 403 to us, which means the configuration is not something we can describe:

• Whether your submitted work is retained in a comparison repository, and for how long — not documented.
• Whether there is any opt-out — not documented.
• Whether students can self-check a draft before submitting — not documented.
• Whether Turnitin’s AI-writing indicator is switched on for Columbia — not documented.

We want to be precise about that last one, because it is the question students actually ask. Columbia names Turnitin as a similarity service. Nothing we could read says Columbia runs AI-writing detection on submissions, and nothing we could read says it does not. It is an open question, and your instructor or your advising dean can answer it faster than any web page.

If a similarity percentage does come back on your work and you are trying to understand what it means, a similarity score and an AI score are different measurements of different things — see what a Turnitin AI score actually means.

What Columbia tells instructors about detectors

Columbia’s Center for Teaching and Learning gives instructors specific instructions on what to do when they suspect misconduct, and those instructions are useful to you because they describe the evidence Columbia expects:

“Look for evidence from the student’s process (e.g., drafts, citations, explanations, or prior work). Note: AI-detection tools are not foolproof, and it is not recommended to rely exclusively on these resources.”

Read the scope of that carefully, because it is easy to over-read. Columbia is not banning detectors and it is not saying their output is worthless. It is saying two narrower things: they are “not foolproof”, and an instructor should not rely on them exclusively. That is a corroboration requirement, and it comes from a page that dates itself: “Columbia Center for Teaching and Learning. (2026). Promoting Academic Integrity. Columbia University.”

For you, the corroboration requirement is the practical point. If a concern is raised, the thing Columbia’s own guidance directs your instructor to look at is your process — drafts, citations, explanations, prior work. Those are things you can have, and they are worth accumulating before you need them: version history, the notes you wrote from, your reading, an earlier essay in the same voice.

The CTL also frames the underlying problem in its own words: “As agentic and generative AI tools continue to distort the meaning of original thought and academic research, the definition of academic integrity must necessarily evolve.” And its recommended answer is assessment design rather than detection: “Assignments are more AI-resilient when students must share how they did their work, not only submit a final product.”

There is one more line worth knowing, because it tells you how a first conversation is supposed to open. The CTL scripts a non-accusatory approach:

“Be direct but non-accusatory, for example: ‘I noticed a strong similarity in submissions. Can you help me understand how you completed this assignment?’”

If you are asked a question shaped like that, it is an invitation to explain your process, and Columbia’s guidance expects you to be able to. Answer it concretely rather than defensively.

We should say where we stand, since we sell a checking tool. Our published error rates are measured on English and German corpora, and no detector — ours included — produces proof. A score is a signal to be corroborated, which is exactly what Columbia’s guidance says. Background: how accurate AI detectors really are and, if you write in English as a second language, AI detectors and non-native speakers.

Dean’s Discipline and the standard of proof

The process is described in one paragraph, and it is a dense one:

“Overseen by Student Conduct, a unit within the Center for Student Success and Intervention (CSSI), the Dean’s Discipline process is an educational one that determines responsibility using the principle of ‘preponderance of evidence.’A student charged with a violation of academic integrity is notified and provided the opportunity to respond through the Dean’s Discipline process with the support of their Advising Dean. If the student is found responsible for a violation sanctions will be issued upon consideration of the specifics of the case, institutional precedent, disciplinary history, aggravating circumstances, and community impact.”

What that gives you, concretely:

The standard is preponderance of evidence. More likely than not. It is a low threshold compared with anything criminal, and it is worth knowing before you assume that an inconclusive detector output cannot support a finding — which is precisely why the corroborating process evidence in the previous section matters so much.

The decider is Dean’s Discipline, not a student honour council. Despite the Honor Code being a student-adopted pledge, the adjudication is administrative.

You have a named support person: your Advising Dean. Columbia writes that into the description of the process. Contact them early — before your meeting, not after it.

Sanctioning is comparative.“Institutional precedent” and “disciplinary history” are listed as inputs, so what happened in similar cases and what is already on your record both feed the outcome. A first incident and a second are not weighed the same.

The appeal route is not publishedon the pages we could read. We are not going to invent one. Ask your Advising Dean what appeal rights attach to a Dean’s Discipline outcome, and ask before the outcome, so you know the deadline.

If a meeting is scheduled

Everything above is easier to read when nothing is happening. If something is happening, here is what Columbia’s own text entitles you to and expects of you.

You are notified, and you get to respond.The policy says a charged student “is notified and provided the opportunity to respond through the Dean’s Discipline process”. That opportunity is the centre of the process. It is not a formality to get through; it is where your account of how the work was produced enters the record.

You have an Advising Dean.Columbia names them in the same sentence — the response happens “with the support of their Advising Dean”. Contact them before the meeting rather than after it. They know the process, they know what institutional precedent looks like, and they are the right person to ask the question this page cannot answer: what appeal rights attach to the outcome, and by when.

The process is described as educational.Columbia calls Dean’s Discipline “an educational one”, and a student found responsible is “referred to meet with the Director of Academic Integrity to reflect on the process and reconnect to campus support resources”. That framing is not a promise of leniency, but it does tell you the room is not a courtroom. A candid, specific account of what you did lands better there than a defensive one.

The first conversation may not be the process at all. The CTL asks instructors to open non-accusatorily — “Can you help me understand how you completed this assignment?” — so an instructor who asks you that may be trying to resolve a doubt, not building a case. Treat it as the genuine question it is. Walk them through your process concretely: where the idea came from, which sources you read in which order, what you cut and why, what you found hard.

Do not overstate anything.This is the one hard warning Columbia writes down: “dishonesty during the Dean’s Discipline hearing process may result in more serious consequences”. If you used a tool, say so and say exactly what for. A use you disclose is a fact in the case; a use that emerges later is a second problem stacked on the first.

Bring your evidence, not your indignation.Sanctioning weighs “the specifics of the case, institutional precedent, disciplinary history, aggravating circumstances, and community impact”. Drafts, notes, version history and earlier work in your own voice speak to the specifics. Nothing else you can bring will.

Step-by-step guidance for the days before a meeting like this is in what to do when you are accused of using AI.

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.

Sanctions, transcripts and what follows you

The range, in Columbia’s own words:

“A student found responsible for an academic integrity violation may expect to receive sanctions ranging from conditional disciplinary probation, disciplinary probation and may even be suspended or expelled from the University. Students will also be referred to meet with the Director of Academic Integrity to reflect on the process and reconnect to campus support resources.”

Note that the floor is conditional disciplinary probation, not a warning. Columbia states the same range more bluntly elsewhere: “If found responsible, violations range from conditional disciplinary probation through expulsion from the university.” And the Faculty Statement puts it more bluntly still: “Students failing to meet these responsibilities should anticipate being asked to leave Columbia.”

On transcripts, the rule is narrow and discretionary:

“The University reserves the right to indicate disciplinary suspension or expulsionon a student’s academic transcript.”

So notation is reserved for the top two sanctions, and even there it is a right reserved rather than an automatic consequence. Probation carries no stated notation. That is better news than many students fear — but read the next paragraph before relaxing, because the transcript is not the only place a finding shows up:

“Additionally, students found responsible for violations of academic integrity may be required to report such offenses on future applications to graduate and professional schools including conditional disciplinary probation. Both academic and disciplinary standings are taken into consideration when awarding College honors. The parents or guardians of dependent students may be notified when a student is no longer in good disciplinary standing. Furthermore, the instructor of record reserves the right to determine how the student’s final grade will be affected in the course where the academic integrity violation occurred.”

Four separate consequences in one paragraph, and the first is the one that lasts longest. Graduate and professional school applications routinely ask whether you have been found responsible for academic misconduct, and Columbia says conditional disciplinary probation — the bottom of its scale — can be reportable. A sanction with no transcript notation can still be something you have to declare for years.

College honours are affected. Parents or guardians of dependent students may be told. And the grade sits with the instructor, separately from the disciplinary sanction.

There is one more, easy to overlook: “Students that are suspended or expelled for disciplinary reasons lose all access to the Center for Career Education’s services.”

Whether a degree can be revoked is not documented on the pages we could read, and we will not assert it either way.

The gap in Columbia’s own guidance

Something worth knowing if you are trying to learn the rules from Columbia’s student-facing material rather than from the policy.

Columbia publishes a long, detailed explanatory taxonomy of academic dishonesty — plagiarism, cheating, fabrication, unauthorised sources of assistance, assisting others, and “other forms”, each with examples. That page contains zero occurrences of “artificial intelligence”, “generative”, “ChatGPT”, “AI”, “Turnitin” or “detect”. Its closest examples remain human and commercial: “Pay someone to write a paper for you”, “Allow someone else to write a paper for you”, “Purchase a pre-written paper”.

In other words, Columbia’s educational material has not caught up with its own violation schedule. The AI rule is real and enforceable; it just is not explained on the page most likely to be read by a student trying to understand what counts.

The practical lesson: read the policy page with the enumerated violations, not only the friendly explainer. And do not conclude from the explainer’s silence that AI is unaddressed at Columbia. It is addressed, by name, in the instrument that matters.

What we could not read

Columbia is fragmented across many hosts, and several of them blocked automated access. Blocked is not absent — these documents are published, and are readable from a normal browser. We just could not retrieve them, so we will not describe them.

• The university-level AI policy at the Office of the Senior Vice Provost. The CTL page links to it, so it exists; the host returned HTTP 403. Its content is unknown to us, and it may well govern things this page does not cover.
• The CUIT Turnitin service page — hence the four unanswered configuration questions above.
• GSAS, SPS and the professional schools, plus Columbia’s essential-policies host and the Engineering site. Those schools’ own rules are not represented here at all.
• Any published case data. No annual figure for allegations or findings was retrievable, and we will not quote one.

Before you submit

• Treat a silent syllabus as a prohibition.The rule is “absent a clear statement… prohibited”, so silence is a no, not a maybe.
• Get permission in writing, per course.A “clear statement” from your instructor is the thing that makes use authorised, and an email reply is a clear statement you can produce later.
• Ask whether early-stage use counts.The rule says “to complete an assignment or exam” and does not carve out outlining or brainstorming.
• Keep drafts, citations and prior work.Those are the four kinds of evidence Columbia’s own guidance directs instructors to look at.
• Remember that intent is not required. An accidental violation is still a violation; care at the writing stage is worth more than an explanation afterwards.
• Contact your Advising Dean earlyif a concern is raised. Columbia names them as your support in the Dean’s Discipline process, and the appeal route is not published — so ask them what it is.

If you want a read on your own writing before you hand it in, our AI check is free up to 1,500 words, with no account and no name. It is a second opinion, not a clearance — and at Columbia the thing that answers a question about your work is your process, not a percentage.

Sources

• Columbia College, Academic Integrity — the Honor Code as resolved by the Columbia College Student Council, the student affirmation, and the Faculty Statement including “student, scholar, or internet agent”. No effective or last-revised date; footer © 2026.
• Columbia College / Columbia Engineering, Academic Integrity policy — the enumerated violation “Unauthorized Use of Artificial Intelligence”, the sanction range, the transcript reservation, the collateral consequences, the Dean’s Discipline description, the preponderance-of-evidence standard, and “intentional or unintentional”. Undated.
• Columbia College / Columbia Engineering, Academic Integrity — the sanction range in short form. Undated.
• Columbia College / Columbia Engineering, Dishonesty in academic work — the long-form taxonomy, which contains no mention of AI, Turnitin or detection. Undated.
• Columbia Center for Teaching and Learning, Promoting Academic Integrity — “CUIT: Turnitin Services”, the detection caution, the process-evidence recommendation, the non-accusatory script and the syllabus checklist. Self-cited as “Columbia Center for Teaching and Learning. (2026).”
• Columbia University Office of the Senior Vice Provost, AI policy, and the CUIT Turnitin service page. Not retrievable by us at the time of writing — both hosts returned HTTP 403, so the documents exist and we could not read them.

This page is orientation, not legal advice, and it covers Columbia College and Columbia Engineering (SEAS) only. What binds you is the Columbia College / SEAS academic integrity policy, Columbia’s university-level policies including its AI policy, and the AI rule written in your own course syllabus.