What Cornell's New Academic Integrity Code Says About AI, Fake Citations and Proof

The short answer. Cornell's revised Code of Academic Integrity, effective 24 August 2026, treats AI help the instructor has not explicitly permitted as unauthorised assistance, makes students responsible for every citation they submit however it was produced, and requires a violation to be proved to a clear and convincing standard. The code does not mention detection tools at all.

Most university AI rules answer one question: is it allowed? Cornell's revised code, which the Office of the Dean of Faculty dates as updated on 27 July 2026 and effective from 24 August 2026, answers three. What counts as unauthorised AI help. Who answers for a citation a chatbot made up. How strongly a violation has to be proved before anything happens to the student. Those are separate questions, and keeping them separate is the useful thing about the document.

This post sets out what the code says, quoting it directly, and what follows for a student or an instructor. These are Cornell's rules. They are not a national standard, and other universities have reached different positions.

What counts as unauthorised AI assistance at Cornell?

The code's definition of unauthorised assistance now names the technology: "Unauthorized assistance includes any aid not explicitly permitted by the instructor, whether provided by notes or reference aids, by other individuals, by artificial intelligence, or by devices such as phones, laptops, or wearable technology."

Two words carry the weight. Assistance has to be "explicitly permitted" by the instructor. Under this wording, a syllabus that says nothing about AI has not permitted it. That is the same logic UBC and Manchester arrived at from the other direction, by requiring instructors to state their AI position on every course; we cover both in is AI allowed in my assignment. At Cornell the default sits with the student: if the permission is not written down, assume it has not been given, and ask.

Who is responsible for a citation a chatbot invented?

Chatbots produce references that do not exist. Cornell's code puts that squarely on the student. Among prohibited conduct it lists: "Citing fabricated or non-existent sources, including false citations generated by artificial intelligence or other tools. Students are responsible for the accuracy and existence of all sources they cite, regardless of how those sources were generated or obtained."

This is the one AI failure that leaves a checkable trace. A reader can look a reference up. Where a detector produces a probability, a fabricated citation produces a fact: the paper exists or it does not. Our piece on whether teachers can tell if you used ChatGPT already lists a citation check among the things instructors actually do, and Cornell has now written the consequence into its code.

How strongly does a violation have to be proved?

The code states the standard: academic integrity violations are evaluated against the "clear and convincing" standard of proof, which it describes as a quantum of evidence beyond a mere preponderance but below "beyond a reasonable doubt". In the code's words, "Clear and convincing evidence will produce in the mind of the trier of fact a firm belief as to the facts sought to be established."

The code does not say what a detector result is worth against that sentence, because it does not mention detectors. Our own reading, stated as ours: an AI-pattern score is a statement that a piece of text shares statistical features with AI-generated writing at a chosen threshold. It is not a fact about who wrote the text, and every detector, ours included, wrongly flags some human writing; our current model flags 3 of 400 real student essays at the flag threshold, a figure published with its method in our study on how often AI detectors flag human writing. A screening result can start an inquiry. Whether the evidence as a whole produces "a firm belief as to the facts" is a question for the people running the process.

A New York court reached the same conclusion from outside the university in January 2026, annulling an Adelphi University finding that rested on a Turnitin score of 100 per cent because of what the university did not do with it; the case is covered in what a court needed beyond a 100 per cent AI score.

Does Cornell require instructors to look at the student's process?

It allows it rather than requiring it. The code says that as part of the investigation, "faculty members may discuss the apparent violation with the student, including questions about the student's process, sources of information, and techniques used in completing the work." May, not must. The code does not mention detection software at all, and it does not restrict what may be used as evidence.

That places Cornell between two positions we have written about. The University of Victoria prohibits instructors from using AI tools to decide whether a violation occurred and bars them from investigations as evidence, covered in can my university use an AI detector to accuse me of cheating. Mount Royal University recommended its staff avoid detectors altogether, covered in why a university told staff to stop using AI detectors. Cornell says nothing about the tools and instead states the standard the evidence as a whole has to meet. The code does not rule on what a detector result counts for; on our reading, a firm belief about the facts is not something a score can supply on its own.

What is the Accepting Responsibility route?

The revised code also folds in a university-wide programme for minor, first-time violations. According to the Cornell Chronicle of 12 August 2026, the Accepting Responsibility programme is an alternative to a primary hearing, which the Chronicle describes as a complex and sometimes adversarial process, for cases such as cheating on part of an assignment or using AI to answer a question in a way that breaks class policy. Students take part in a workshop on values and habits and accept reduced assignment penalties. The code page summarises the route as accepting responsibility without a formal finding, completing an educational workshop, accepting limited penalties and waiving appeal rights.

The Chronicle reports a two-and-a-half-year pilot from spring 2024 involving about 360 students across 100 courses, with four students recording a second violation, and says more than half of the violations during the pilot were AI-related. The Faculty Senate updated the code to include the programme, effective 24 August. Liz Karns, Cornell's director of academic integrity initiatives, is quoted on why: "Primary hearings take a substantial amount of time", and "This can be disruptive to the educational process".

What should a student do under rules like Cornell's?

Find the permission in writing. Under Cornell's definition, help that is not explicitly permitted is unauthorised. If your syllabus is silent, ask before you use anything, and keep the answer.

Check every reference exists. Open each source you cite. A citation a tool produced is your citation the moment you submit it.

Keep your process. Drafts, outlines, version history, notes of conversations about the work. Cornell's investigation may ask about your process and sources; being able to answer is the strongest position you can be in. A university course that built its whole assessment around this kind of record is described in can assessment design replace AI detection.

Know the standard. If you are told a detector flagged your work, the question Cornell's code asks is whether the evidence as a whole produces a firm belief about what happened. Ask what that evidence is.

What should an instructor take from it?

State what is permitted, in the syllabus, before the first assignment. Check citations, because that is the check with a definite answer. Use a screening result as a reason to look more closely and to ask about process, never as the finding. Under a clear and convincing standard, a conversation about how the work was produced is, on our reading, worth more than any score.

Who wrote this, and what we sell. Is It AI is an AI-writing detector for teachers and students. Our result is an AI-pattern score, a screening result, and our full published figures, including the cases we miss, are on the methodology page. All quotations from Cornell's code and the Cornell Chronicle are from the pages linked below. We have no connection to Cornell University.

Sources

• Cornell University, Office of the Dean of Faculty, Code of Academic Integrity effective August 24, 2026, updated 27 July 2026. All code quotations are from this page.

• Cornell Chronicle, Holly Hartigan, New program turns minor cheating violations into a learning opportunity, 12 August 2026.

• Supreme Court of New York, Nassau County, Matter of Newby v Adelphi University, 2026 NY Slip Op 26021, decided 28 January 2026.