AI Cheating Cases Are Up 400%, and Faculty Detection Is Worse Than a Coin Flip
Academic-integrity cases at Chapman University are up 400% since the pandemic, and fewer than 50% of faculty can reliably tell AI-written text from human writing. The four-year degree is the thing being wagered.
Key Takeways
- Academic-integrity cases at Chapman University are up 400% since the pandemic, and Grinnell College reports a 170% increase over the last 3 academic years (The Chronicle of Higher Education, August 2026).
- University of Calgary research finds fewer than 50% of faculty can reliably tell AI-written text from human writing, which puts campus detection below the accuracy of a coin flip.
- 60% of college students use AI weekly for coursework, according to an April 2026 Lumina Foundation and Gallup poll, while more than 80% of faculty surveyed by The Chronicle say they have redesigned assignments to resist AI.
- A degree is a 4-year wager that assessment can verify what a transcript claims, and AI cheating volumes are testing that verification faster than most review committees can be staffed.
65% of faculty surveyed by The Chronicle of Higher Education say they have already caught a student using AI to cheat. University of Calgary research puts the share of faculty who can reliably tell AI-written text from human writing at under 50%. Both numbers describe the same people, in the same semester, doing the same job. AI cheating is being caught at scale by people who cannot verify it at scale.
That arithmetic is the actual story in the Chronicle's August 13 reporting, and it is showing up on campuses that have little else in common.
Case volumes are rising faster than the process built to review them
Academic-integrity cases at Chapman University are up 400% since the pandemic. Lindsay Waldrop chairs the process that reviews them. Every one of those cases lands on a real desk as a person, a semester, and a decision that follows a 19-year-old for years. At Grinnell College, cases are up 170% over the last 3 academic years, and Andrea L. Tracy chairs the committee that hears them.
| Institution | Increase in academic-integrity cases | Period | Chairs the review |
|---|---|---|---|
| Chapman University | 400% | Since the pandemic | Lindsay Waldrop |
| Grinnell College | 170% | Last 3 academic years | Andrea L. Tracy |
Neither number describes a failing school. Both describe a review process designed for a steady trickle that is now handling a flood. Waldrop said the part most administrators would edit out of a quote:
I didn't have a lot of answers for that as chair of that committee, because I don't think there are a lot of answers right now.
That is candor from someone standing inside the problem, and it deserves better than a scoreboard.
Detection was the first part to break
More than 80% of the faculty in that same Chronicle poll say they have tried to AI-proof their assignments. Andrew Peck, an assistant dean at Penn State, named the real difficulty: "AI has really forced a lot of nuance. Using a prohibited tool or a tool in a prohibited way isn't new. What's really new is that it's very, very hard to detect."
Detection used to be the cheap part of academic integrity. It stopped being cheap. And the people who have to make the call are being asked to do it against a base rate worse than a coin flip, which is what the Calgary work from Sarah Elaine Eaton actually measures.
Meanwhile, 60% of college students use AI weekly for coursework, according to the Lumina Foundation and Gallup poll from April 2026. Students are doing what people have always done with a tool that is free, fast, and sitting on every phone. Adoption ran ahead of policy, the same way it does inside every organization I advise, and the same way it is running right now inside the university systems where I sit on the governance side of the table. The students are ahead of the institution here, and treating that as a character flaw burns time the institution does not have.
The degree is what is actually being underwritten
A degree is a bet placed over 4 years. The student wagers time and tuition. The institution wagers its name on a claim that the person holding the diploma can do what the transcript says. Call it the four-year bet, and understand that assessment is the only place that bet gets checked before it goes out into the world.
Most of the conversation about AI in education focuses on curriculum half-life, the speed at which what gets taught stops matching what the work requires. That is a real problem and a slow one. The faster problem is that the verification method has a half-life too, and right now it is the shorter of the two.
A degree is a promise about what a person can do. Assessment is where that promise gets checked.
Three questions worth more than another detection tool
These are the questions I am carrying into my own advisory work this fall, and I do not have clean answers to all of them either.
How much of a student's grade rests on work an instructor actually watched happen? Oral defenses, in-class writing, versioned drafts, and lab practicals verify something a submitted file cannot. Every one of them costs faculty time, which is the binding constraint, and any plan that pretends otherwise dies in week 3.
Does the campus have a disclosure rule written down, per assignment, in language a 19-year-old can follow on the first read? Grinnell starts there. Tracy put it plainly: "If they are disclosing their sources, that's not an academic-integrity violation." A rule students can read beats a detector faculty cannot verify. Florida wrote one rulebook across 28 public colleges rather than leaving 28 campuses to draft 28 versions. The CU System paired its ChatGPT Edu rollout with AI literacy instead of a prohibition nobody could police. Both chose to teach the rule.
And who is staffing the review process at 4 times the old volume? Chapman's caseload multiplied. The committee did not. Staffing is the budget question that rarely makes it onto the strategic plan, and it is the one that decides whether any policy survives contact with a real semester.
An employer reading a transcript is reading a claim that somebody checked
This runs well past campus. A hiring manager who stops believing the transcript starts running their own assessment, which is slower, more arbitrary, and much easier to get wrong than the one the university was built to run. Families feel it first. Watch any parent writing a tuition check this fall and doing the math on what the credential still buys.
4 years is a long time to hold a position on a technology that shifts every 6 months. That is the wager every institution makes on behalf of every student it admits, and the honest work happening right now is the work of making the wager good.
Faculty are not behind on this. They are several years into a problem with no settled answer anywhere in the country, and the ones worth listening to say so plainly. What moves it forward is a decision your campus can make this term: what share of a degree gets verified in a way AI cannot stand in for, and who gets paid to do that verifying.
Put a number on it at your next faculty meeting and watch what the number does to the conversation. I am easy to find if you want to compare notes.
Can professors tell if you used AI to write an essay?
Often they cannot verify it, even when they suspect it. University of Calgary research puts faculty accuracy at under 50%, and Williams College committee chair Daniel Barowy told The Chronicle his committee could not independently verify what detection tools claimed, because they did not have time to engage in the scholarship. Most cases now turn on process evidence such as drafts, version history, and a conversation with the student, rather than on a detector score.
What should I ask a college about its AI policy before enrolling?
Ask what share of grades comes from work an instructor watches happen, whether the AI disclosure rule is written per assignment rather than per campus, and how many people staff the academic-integrity review process relative to current caseloads. Those answers describe how a degree will hold up far better than any statement of academic values.
Why are AI cheating cases rising even at colleges that permit AI use?
Permission is usually set per assignment, so a tool that is allowed in one course becomes a violation in the next, and students carry the burden of tracking the difference. Penn State assistant dean Andrew Peck described this as the nuance AI forced into a process built for clearer rule-breaking. Volume rises because ambiguity produces referrals on its own, separate from any change in student behavior.
Here is what makes Alex a credible voice on this topic: Alex advises the California State University system on AI governance and is Innovator-in-Residence at Tulane University, which puts him inside the same assessment and academic-integrity decisions this piece describes, on the committee side of the table.
If your institution is rebuilding how it verifies student work, book a conversation →
.svg)
