Don’t miss anything! Sign up for my newsletter & join our Facebook community of 1,000+ like minded thinkers!

AI Cheating in College Is a Symptom. Our Assessments Are the Disease.

Last spring an economics professor at Brown University gave his students a take-home midterm. In past years the class average had landed somewhere between 65 and 80. This time it was 96. Forty of eighty-six students earned a perfect score.

Professor Roberto Serrano did what any suspicious professor would do. He gave the exam to ChatGPT. The machine's answers were technically correct and oddly convoluted, and most of the students' answers matched them. He moved the final exam into a classroom with pen and paper. About a third of the class dropped or skipped it. The average on the in-person final was 48.

It is tempting to read that story as a morality tale about lazy students. I think that misses the point. The Brown midterm did not reveal that students had suddenly become dishonest. It revealed that the take-home exam had stopped measuring what it was designed to measure. The tool changed. The assessment did not.

How Big Is the Problem, Really?

Faculty certainly believe it is big. In a national survey released in early 2026 by the American Association of Colleges and Universities (AAC&U) and Elon University's Imagining the Digital Future Center, more than a thousand U.S. faculty weighed in. Seventy-eight percent said cheating on their campus had increased since generative AI became widely available. Ninety percent said they expected AI to weaken students' critical thinking. Nearly three in four said AI threatens the integrity and value of a degree.

The student data paints a more complicated picture. In May 2026 researchers René Kizilcec of Cornell, Igor Chirikov of UC Berkeley and Ivan Smirnov of the University of Technology Sydney published a study in Science drawing on survey responses from more than 95,000 students at twenty U.S. public research universities. About a third regularly used generative AI on assignments. Using a technique designed to get honest answers about sensitive behavior, they estimated that about 9 percent used it to cheat, lower than the horror stories suggest. But among daily users, the cheating rate rose to 26 percent.

Their conclusion was blunt. In Kizilcec's words, "Assessment reform is necessary and urgent."

So both things are true. Most students are not cheating. And the students who lean on AI most heavily are cheating at rates that make many of our traditional assignments unreliable.

We Cannot Even Agree on What Cheating Is

Here is the finding from the AAC&U survey that should worry every dean. Faculty do not agree on where the line is.

Asked whether using a detailed AI-generated outline counts as cheating, 52 percent of faculty said yes, 29 percent said no and 18 percent were unsure. Asked whether asking AI to edit a paper the student wrote counts as cheating, 34 percent said yes and 45 percent said no.

Think about what that means for a nineteen-year-old taking four courses. The same behavior can be encouraged in one class, forbidden in the next and undefined in the third. Academic integrity councils, as the Chronicle of Higher Education has reported, are struggling to keep up. Some students have even sued over accusations based on unreliable AI detectors.

When the rules are unclear, integrity becomes a guessing game. That is not a student character problem. It is an institutional design problem.

The Philosophical Mistake Behind the Panic

As a philosophy professor, I want to name the deeper confusion. For a century we have treated the product of student work, the essay, the problem set, the lab report, as reliable evidence of the process of learning. That was always an inference. It worked because producing the product required the process.

Generative AI broke the inference. A student can now produce the artifact without undergoing the thinking. The essay is no longer proof of the mind behind it.

Socrates would not have been surprised. In Plato's Phaedrus, he worries that writing itself would give students the appearance of wisdom without the reality, because they would rely on marks outside themselves instead of understanding within. Every new tool of thought revives that worry. The answer has never been to ban the tool. It has been to change what we ask students to show us.

What Is Working in Real Classrooms

The Chronicle has documented how professors are adapting, and a few strategies keep appearing because they work.

  • Bring key thinking into the room. In-class writing, oral exams and handwritten work are making a comeback. The Cornell team explicitly recommends a return to some controlled, in-person assessment.
  • Grade the reasoning, not just the result. San Jose State chemistry professor Gianmarc Grazioli weights 30 percent of project grades on students explaining their own work. If you cannot explain it, you do not own it.
  • Make the process visible. Some instructors require shared documents with version history, process journals or brief reflections on how a student reached a conclusion.
  • Design for the tool, openly. Nearly half of the professors responding to a Chronicle reader survey had introduced assignments that expect students to work with AI and then critique, verify or improve its output. That turns AI from a shortcut into a sparring partner.
  • Lower the stakes where possible. Northwestern computer scientist Edith Elkind asks students to disclose their AI use on homework and grades for completion, arguing that fear of a bad grade drives much of the cheating.

A Simple Framework for Faculty

When faculty ask me how to redesign an assignment, I offer three questions.

  1. What thinking do I actually want to see? Name it precisely: weighing evidence, constructing an argument, applying a concept to a new case.
  2. Where can I watch that thinking happen? In conversation, in class, in drafts, in a defense of the work. If the only evidence is a finished file, the assignment is vulnerable.
  3. What is the honest role of AI here? Forbidden, permitted with disclosure or required as a tool to critique. Say it in writing, every time.

None of this requires becoming a detective. It requires becoming a better designer of learning.

The Opportunity Inside the Crisis

I teach at a community college in Chicago, and I care about this for a reason beyond integrity policies. Our students need degrees that mean something. If employers come to believe that a transcript only proves access to a chatbot, the students hurt most will be the ones who can least afford a devalued credential.

AI cheating is a symptom. The disease is an assessment model built for a world where producing work required doing the thinking. That world is gone. The good news is that the fix points us back to what education was always supposed to be: not the production of artifacts, but the formation of minds that can reason, explain and judge.

If we redesign well, AI may end up forcing the most important reform in higher education in a generation.

Sources and Further Reading

Bringing this conversation to your campus? Dr. Sweis leads faculty development workshops on teaching thinking in the age of AI. Learn more about workshops and keynotes.

You Might Like These