Academic Integrity Councils Buckle as AI Cheating Cases Surge

A system built for plagiarism meets a machine that leaves no trace

For decades, academic-integrity cases worked in a predictable way. A professor suspected a student of copying, put the paper through a plagiarism checker, and took the match to a small panel of faculty and sometimes students. Generative AI has broken that process. The reason is not only that it makes cheating easy. It also leaves the people judging a case with very little they can prove.

The numbers from individual campuses are striking. Chapman University says academic-dishonesty complaints from faculty have risen 400 percent since 2020, with ChatGPT's public release in fall 2022 falling inside that period.1 Grinnell College's Committee on Academic Standing reports a 170 percent increase in integrity cases over three academic years.2 At Pennsylvania State University, the growth is hard to measure because professors have started filing complaints against several students on one form.2 Andrew Peck, Penn State's first dean of academic integrity, called the volume "overwhelming."1

Colleges are responding in different ways, but the direction is the same: less formal punishment, more prevention and redesign. The question in the headline, whether AI has made the integrity council obsolete, can be answered fairly directly. The council as a hearing body built around evidence of copying is wearing out. What is replacing it is less a tribunal than a set of decisions about how courses are taught and graded.

The proof problem

The central issue is evidence. AI-generated text is new text, even if no person wrote it, so it never matches a source the way copied work does.1 Detection software can point to stylistic "tells," but it cannot prove anything. Penn State no longer accepts detector results as evidence in integrity proceedings, because a student can simply answer one detector's verdict with another detector's opposite verdict.2

Human judgment has not done much better. Sarah Elaine Eaton, who runs the Postplagiarism Research Lab at the University of Calgary, says a 2023 campus survey found that fewer than half of faculty could tell human writing from AI writing, which she called "worse than a coin toss."2 She considers it "a futile task" to ask an honor board to decide whether AI wrote an essay.2

The result is friction between faculty and the panels that hear their complaints. Integrity committees at Grinnell and Williams College use a "preponderance of evidence" standard. When the evidence is uncertain, Williams sides with the student, and its chair, Daniel Barowy, says this has left some professors "incensed."2 Panels also often lack the subject expertise to weigh what faculty bring them. Barowy described a statistical argument about identical problem-set answers that the committee could not evaluate, and a Beowulf case it could not check without doing the scholarship itself.2

The wider sector has turned against detection for similar reasons. Yale, Vanderbilt, Johns Hopkins and Indiana now ban or discourage relying on detectors as the only evidence of cheating, and at least a dozen universities, including Northwestern, Georgetown and NYU, have turned off Turnitin's AI detection feature.21 Studies have found that these tools produce false positives and are biased against non-native English writers, and some wrongly accused students have sued.21 Harvard College Dean David Deming told students that everyone "would all benefit from getting out of the AI-detection business because it damages trust between faculty and students."23

How campuses are rewriting the rules

The policy responses fall roughly into three groups.

Redefinition. Grinnell treats AI output as a source like any other. Students commit an academic-honesty violation only if they fail to cite it. A student who uses AI where a professor has banned it, but cites it, has broken a course policy, which is a lesser offense the professor handles with grades.2 Chapman is discussing a policy that would allow AI by default unless a professor says otherwise. The committee's chair, Lindsay Waldrop, expects it to cause "chaos" over where cheating begins.2

Restructuring. Penn State is closing the integrity councils on its 14 University College campuses and moving cases to a central office, hoping to free campus resources for prevention.12 Grinnell is ending conduct hearings altogether. A designated official will instead hold facilitated conversations between student and professor and decide responsibility and any sanctions, an approach chair Andrea Tracy hopes will be less punitive and quicker.2

Retrenchment. Some institutions are tightening rules instead. Berkeley Law's policy forbids AI for outlining, drafting, editing, translating and exams unless an instructor makes an exception.26 Princeton's faculty voted to require proctors for all in-person exams.25

These approaches pull in opposite directions, and that matters. Students already face rules that change from one course to the next. An Associated Press report found that schools mostly leave AI policy to individual teachers, so rules vary widely within the same building, and students avoid asking for clarification because admitting any AI use could mark them as cheaters.5 Eaton notes that most North American policies still do not define AI misuse at all, which leaves institutions stretching old plagiarism definitions over a new kind of misconduct.2

How much cheating is happening

The coverage disagrees most on scale, and the disagreement shapes how campuses respond.

The largest dataset is somewhat reassuring. A study in Science surveyed 95,500 students at public research universities and found that two-thirds used generative AI in 2023-24, but only 9 percent admitted knowingly submitting AI work where it may not have been allowed. Among daily users, though, that figure rose to 26 percent, which co-author René Kizilcec reads as a sign that cheating will grow as use grows.

Other data is bleaker. In a Harvard Crimson survey, one-third of the graduating class admitted using AI against instructor permission. Ninety-three percent of those students said they were never caught, and only 2.6 percent said they faced any consequence.25 The Atlantic's essay on these numbers argues that students are learning that integrity is "for suckers."25 Faculty sentiment matches that alarm: in a survey by AAC&U and Elon University, 78 percent of faculty said cheating had increased since generative AI arrived, and 73 percent said they had dealt with an AI-related case themselves.23

These figures do not actually contradict each other. They measure different things: self-reported violations in a large sample of 2023-24 students versus outcomes at one elite campus a couple of years later, when tools were better and more common. Taken together, they suggest that today's 2026 figure is higher than the Science baseline and that very few cases are being enforced. A rule that is almost never enforced is not much of a deterrent, which is one reason the shift toward assessment redesign looks like a necessity rather than a fashion.

The tutoring evidence complicates the picture

The integrity debate usually treats AI as a threat to learning. The research on AI tutoring shows that the design of the tool largely decides whether it helps or hurts.

The best-known positive result comes from Harvard. In a crossover study, 194 physics undergraduates using a custom-built AI tutor achieved more than twice the learning gains of the same students in an active-learning classroom, and spent less time doing it.19 The researchers stress that the tutor was built on established teaching principles and was not simply ChatGPT, and that the study covered only two lessons at one university.19 Two randomized trials described by Stanford's National Student Support Accelerator found that AI built into live math tutoring improved outcomes. Students using Google's LearnLM did better on later, harder topics than students tutored only by humans, and the Tutor CoPilot tool helped less-experienced tutors the most.11

A new working paper points the other way. In a randomized trial with 2,379 undergraduates and 30 instructors at a large public university, access to a course-integrated AI tutor lowered final grades by up to 0.37 standard deviations. It also sharply reduced participation in instructor-designed online activities, and first-generation students lost more than twice as much as their continuing-generation peers.13 The authors also cite earlier work showing that general-purpose AI access hurt later independent performance in high school math, while tutoring guardrails reduced that harm.13

This is where the tutoring research and the integrity debate meet. In both, the underlying risk is that students hand off the thinking. A well-designed tutor makes students do the reasoning. An unstructured chatbot, or a homework task a chatbot can finish, lets them skip it.

The reading: the essay is under more pressure than the council

The council is not going away entirely. Cases involving fabricated sources and other old forms of misconduct are still common. Notre Dame's integrity director says made-up citations count as misconduct whether or not AI produced them.28 But the expectation that a faculty panel can rule on whether a machine wrote a given essay is no longer realistic, and the institutions moving fastest seem to know it.

The experts quoted most often agree on the alternative. Tricia Bertram Gallant of UC San Diego says policing AI use is no longer a workable way to build integrity.21 Eaton goes further and argues that the take-home essay should be allowed to "die with dignity."2 Professors are turning to oral exams, AI-use logs and process-based grading.23 Getting most faculty to change how they assess students, as one University of Mississippi lecturer put it, will be "the biggest lift."21

That change carries its own risks. If redesign means a broad return to blue books and proctored rooms, it could exclude the online and nontraditional students whom flexible formats were meant to serve, as one WCET analysis warns.9 The working-paper finding that first-generation students were hurt most by an AI tutor is a reminder that equity questions apply to both the threat and the fixes.13 The integrity council's declining role is mostly a sign of a larger shift: colleges now have to show what students have learned, rather than just catching the ones who cheat.