Is it a surprise that students are using AI to cheat when they get the opportunity? I guess it does surprise some people, for a pretty clear case of cheating at Brown University made headlines at Inside Higher Ed (click on the screenshot to read).
Excerpts:
For the first time since he started teaching Welfare Economics and Social Choice Theory nearly two decades ago, Brown University economics professor Roberto Serrano gave his students a take-home midterm this spring. Quite a few students had expressed anxiety about being in a classroom after a gunman killed two students and injured nine in a December mass shooting at Brown, and so “it was appropriate,” he said, to allow students to take their exams at home.
But by the end of the semester, Serrano regretted the decision. Dozens of students in the class likely used artificial intelligence to cheat and earn perfect or near-perfect scores on their midterm, he said. Serrano in turn made the final exam in-person, which led more than a dozen students to drop the course and even more to fail it. Administrators’ response to the widespread cheating event has been “meek,” he said, and the incident has raised questions about how universities can—and should—respond to AI-enabled cheating at scale.
His welfare economics class typically attracted up to 30 students, but this spring he taught 86—an increase he attributes to the promised take-home exams. When the midterm came along, the average score was 96 percent.
“Historically the average grade in the midterm of this course has ranged between 65 and 80 [percent], and this exam was harder than the exams I wrote in the past, because … take-home is an opportunity to challenge the class a little bit more, given that you’re giving the students unlimited time,” Serrano said.
He knew something “fishy” was going on, and so he and his graders ran the test through ChatGPT. The AI gave answers that mirrored what his students had written, and which were “kind of correct, but very off and with a very convoluted style,” Serrano said. For example, one question asked students to prove a mathematical statement that could most obviously be done using a “direct argument,” Serrano explained. ChatGPT—and many of his students—used a “contradiction argument,” which gave the right answer but was “very contrived” and which Serrano could tell wasn’t written by a human.
Prof Serrano said he wasn’t going to fail students for cheating (the allegation can’t be proved) but would give them a final exam in class. At that announcement 18 students dropped the course and 9 decided not to take the final.
Here’s the distribution of scores for both exams, with the final in gray and the midterms in orange. They didn’t plot a correlation, but there doesn’t appear to be one because many of the students who did terrible on the final got nearly perfect scores on the midterm:
The University turned a blind eye to this cheating—for that’s what it surely is—until Serrano went public.
In May, Serrano submitted the data shown above to Brown’s Standing Committee on the Academic Code and received no response. After he went public with his story in late June, the committee, through his department chair, asked Serrano to submit individual complaints against each student suspected of cheating, including copies of their exams, he said.
Most people with a sense of right and wrong would find this behavior unethical, tempting though it may be to cheat, but some misguided profs think that, well, if the chance to cheat is there, there’s nothing wrong with taking it. Here’s from P.Z. Myers, who found the same results with his students.
I figured this out back during the pandemic, when by necessity I had to offer exams online. Scores shot up! I knew immediately what was going on, but I didn’t punish the students — I couldn’t blame them for taking advantage of the system.
“I couldn’t blame them for taking advantage of the system”?? That’s equivalent to telling the students, “It’s okay to cheat if you can get away with it.”
More:
Asked about the university’s response to the cheating, Brown spokesperson Brian Clark told Inside Higher Ed that the procedure for investigating cheating allegations is the same whether it’s one student or several.
“Brown treats every allegation of academic integrity with the utmost seriousness. In regard to this economics course, multiple academic leaders from Brown were in touch with the faculty member who raised concerns to provide details about how the allegations raised could be formally adjudicated. To date, the faculty member has not provided the necessary details to the Standing Committee on the Academic Code to pursue this path toward resolution,” Clark wrote.
If Brown took it seriously, why didn’t they respond to Serrano’s original complaint? Now Brown is adjusting its rules—to be easier on the students! My bolding:
As the cheating scandal unfolded in Serrano’s classroom, a Brown committee on generative AI in teaching and learning was examining how the technology was being used at the university and formalizing recommendations for how Brown can adapt and respond to AI developments. Its inaugural report was published Tuesday.
Three-quarters of Brown professors said they are concerned about students using AI to cheat, according to feedback from 105 faculty members. The same share of respondents said the same in a 2025 nationwide survey from the American Association of Colleges and Universities. As part of a set of medium-term recommendations, the committee encouraged the university to amend the College Academic Code and the graduate student code “to address GenAI realities and safeguard against misuse.”
The committee also suggests that faculty “de-emphasize punishment” and avoid highly restrictive rules around generative AI use.
There is no way to check with 100 percent accuracy whether GenAI has been employed, and norms are likely to change in the years to come,” the report states. “Moving the conversation beyond punishment also allows for open dialogue that will be necessary as Brown navigates this moment.
What is motivating this softness on cheaters, which includes relaxing bans on using AI? I can see this only as a misguided extension of the victim/oppressor narrative, with the students seen as victims. Alternatively, it could be the new mentality that sees students as customers of the university, with their high tuition buying them good grades and a degree. ChatGPT, which launched at the end of 2022, was the first program that made AI cheating widely available, and the oppressor/victim dichotomy was intensified by the killing of Geore Gloye in 202.
The problem of course is that how can you prove that a student used AI unless more than one student gives the same answer. I know there are programs to detect whether AI was used, but I also hear they’re not that good.
The only solution is to not let students have take-home exams, and ensure that in class they cannot have their laptops or phones open, or surrending them before you take the test. (I am told by one professor that students got around this by bringing two phones and then cheating on the second phone when they ask to go to the bathroom).
It would be harder to cheat in labs, for you have to actually do something like run experiments or synthesize an organic compound. I remember that in O-chem (the bane of all biology students) we had to extract caffeine from coffee, and then you were tested when the prof looked at the melting point of your caffeine when it was tested in a glass tube. That would be hard to fake with AI, though I suppose you could purchase caffeine. But if professors want to ensure that students don’t cheat, they’ll have to be creative. Recent surveys show that over 60% of college students admitted to cheating in some way, and that goes up to 95% if you consider high school.
****
Here’s a tweet lauding the highest achiever and two students who apparently didn’t cheat as their midterm scores weren’t that different from their final-exam score.
Friends don’t let friends continue to give exams and papers that are not adapted to the current AI landscape (this is AFTER 27 bailed so the reality is even worse)
PS Props to students 1, 22, and 31 pic.twitter.com/3LPj66YuQT
— Brendan Nyhan (@BrendanNyhan on 🟦☁️) (@BrendanNyhan) July 8, 2026
h/t: Luana
I’m curious … I use Grammarly to check my writing; is that cheating?
You properly used a semicolon; this is definitely AI.
If you were taking a grammar class, the answer would be Yes. Otherwise, No as I see it.
I’ll bet that the students don’t regard this as cheating. If you give them a take-home test, what are they allowed to do?
May they consult notes they took in class? May they consult a textbook? Use a calculator? A spellchecker? May they Google for help (which these days gives an AI answer)? Every student from now on will regard using tools like AI as normal, just as they regard using a pocket calculator as normal.
The lesson here is that expecting students not to “cheat” is not going to work. If you want to assess their ability to do tasks unaided then you need to set exam-conditions exams.
Many university administrators will have a hard time accepting that.
I agree…I think that students regard AI as similar to a calculator…
As a non-professor, why would the university admins not accept all grading to be based on in class, closed book exams? That seems to be the only way to determine how much students are learning.
There goes term papers!
The first thought that comes to mind is to require oral defenses of term papers, to confirm that the student wrote and, more importantly, understands it. I realize that would be a lot of work for the professor, though.
For many, many years, this is the process I have used (did use? I pulled the retirement core in june…) teaching engineering and CS. ANY larger assignment– engineering project, software project, Capstone, individual or group, etc– had regular, in person meetings, face-to-face evaluation, and an oral defense. A pain? Sometimes. Feedback from the students post-graduation and into career is strongly positive, no matter how much they bitched at the time. Lots of paper documentation, as well.
Until 2020/2021, I could give take home exams and generally expect honor code compliance, spot checking with oral questioning. Since then, nothing but in-person, well proctored. I have/had no problem telling a class that it wasn’t that I don’t trust them, but that I don’t trust anyone, nor do/did I have any compunction calling out issues– “Bobbi! That is not your paper!”, “Lesley! WTF? You know she has a different version!”.
I’m a (retired) professor, and I agree with you. Were I teaching today, I would sacrifice class lecture time and require all heavily-weighted work to be done in class. They would learn less, but at least they would learn SOMETHING…but also, yes, Jerry’s remark about term papers and student research remains a major issue.
I’m tempted to put that on a poster or plaque:
They learn less, but they learn something.
It’s the academic version of a debt collection agency’s reply to clients’ objections about their substantial fees:
Eighty percent of something is better than 100% of nothing.
Arguments against actual exams: (1) they are stressful for students, who thus dislike them, and what students like and dislike counts for a lot these days. (2) weaker students find them hard, and can’t get around them by methods such as AI, and thus dislike them — and then there’s the big one: (3) “minority” students tend to do less well in them, therefore they are not “equitable”, and in a proper exam it’s way harder to fudge this point (“minority” here means black and Hispanic students; Asians or Jewish students are also minorities, of course, but tend to score fine, so aren’t “minorities”).
Yeah, I’ve made this argument in the past. To get students not to cheat, you have to first convince them your definition of cheating is reasonable.
How is having AI do the exam different from having another person do it for them?
It’s much more accessible. Unlike (say) Henry Ford II who got good quality cheating work from his father’s employees, anyone can get good quality work from an LLM. It’s inclusive, equitable, and serves diverse communities. The scalable democratisation of cheating.
The impetus to cheat stems from our reliance on grades as a measure of learning. Beginning with the first grade, students are conditioned to earn a grade rather than learn. We have made learning nothing more than a game. And, sadly, cheating is a part of the game. “If you’re not cheating, you’re not trying.”
Grades are just a measure of right and wrong answers. At some point a student must be evaluated on what they know. “If you’re not cheating, you’re not cheating.”
…and it is only cheating if you get caught.
Jerry’s right about correlation … none between scores when all students used. Dropping two lows and all 100s produces r = .38, still not great. As expected given r, measure of reliability of two scores (Cronbach’s alpha = .07) is abysmal, meaning average does not even meet a very basic requirement for a quality measure of performance. But at least average will moderate effect of cheating (avg ~72). If allowed (probably not given current environment where students = clients and instructor discretion is worth nothing), I’d change cut-off for a pass with some recognition for students who appear not to have cheated.
Then there will be products and services that enable students to appear not to have cheated. It’s an arms race, and hallowed academic traditions are not winning. R.I.P.
Probably I’ll be told I’m not allowed to do it anymore for some nonsense ed school logic reason, but I’ve moved to oral exams for advanced physics courses for this reason.
PZ is OK with dishonesty?
Quelle surprise.
Brown grades pass/fail, right? How hard it is to pass without cheating? For real schools, though, they need to address cheating or their diplomas will become literally worthless. They are doing their students a disservice to a degree that could almost be characterized as fraud.
Eating the seed corn. Setting fire to the furniture to keep warm. Etc.
Wow. That graphic says it all. Obviously we still need to figure out how to use AI as a tool, rather than as a crutch. Somehow we’ll have to get there.
When I took my first chemistry course in 1974, calculators weren’t allowed. We had to use slide rules. (I was a rare student with a calculator I couldn’t use.) By the time I took my second course, the next semester, calculators were allowed. AI is one freaking amazing calculator. It doesn’t just make a calculation easier: it produces the final answer, the complete mathematical derivation, the history of the problem at hand, where it fits into the advance of civilization, which tectonic plate was where at the time—and it does all of this in compact form with perfect grammar. This is not going to be easy!
Norman,
I might quibble with your use of “the” as in “the final answer”. I would suggest “a” as in “a final answer”. Nobody has any idea whether or not the ai suggested answer is right (the answer) or just its best fit opinion.
I hate ai.
Although I only skimmed the Brown report, two paragraphs caught my eye. The first one was this:
But later in the report, I read this:
So, students who use AI the most express the most concern that AI does NOT help them learn or think, but may actually impair those abilities. Furthermore, even if AI does aid some students in learning, some of the time, it doesn’t provide students with any help that they can’t already get from other sources.
This report just supports my gut feeling that the use of AI by students should be entirely banned from high school through college.
Another benefit would be to avoid the race to the bottom, where students who would rather than not use AI feel forced to use it in order to remain competitive with the students who do.
In any event, here’s a video of another professor – a statistics professor, no less – who caught his class cheating. Or at least he thought he did.
Prohibition is a non-starter. Today’s job market requires LLM skills for many of the (remaining) jobs. The whole academic assessment and credentialing enterprise is undergoing a revolution, like it or not.
Heck. Kids may have taught how to use AI effectively if they want to get a job.
“I couldn’t blame them for taking advantage of the system.”
Academic cheating is only the tip of the proverbial iceberg. In many ways, it’s old hat; we have always had with us those who game the system to their advantage and see those of us who don’t as losers. But gaming the disability system to one’s advantage in the academic world is an entirely different level of shamelessness—and one is most likely to find it in our wealthy zip codes and the Ivy League. While there might be technical solutions to academic cheating, they won’t solve what appears to be an erosion of elite norms.
The meritocracy always had its mythical elements. We’ll someday soon miss those parts that weren’t.
https://www.socialpolicylab.org/post/understanding-the-importance-of-addressing-economic-inequality
IMO, someday is today.
Who would consult a homeopath if they knew that he used AI to cheat in his exams?
Snap!
I saw the decline of academia over the past several years at the university where I worked (now retired). There were several interacting factors: the university administration turning from an education model to a business model where students were customers to be kept happy at all costs; the ever-increasing status of education “experts” over professors, requiring constant monitoring and training of the latter; the forced adoption of an education fad that rejected supervised exams in favor of independent student assignments; and the rise of AI-based cheating. Friends who are professors at other universities tell the same story so this was not unique to my university. If I was a young person today, I would probably choose a different career. The glory days of academia are over. But what concerns me most about these trends is the likely future outcome – a massive decline in general competence and expertise.
All classes need in-class hand-written tests. But students will complain and university managers don’t really care about academic integrity. Problems are amplified in countries such as Australia and New Zealand that cannot afford any measures that might lead students to not enroll.
[Over my posting limit, but I gotta know…]
What makes those places more vulnerable than others?
Competition between many unis for a dwindling pool of potential students. Keep the students you do have happy and they will rate the uni highly in polls (good for advertising) and perhaps even recruit their friends and family, was what I was told.
Is that unique to Australia and NZ?
No. Administrative bloat is another factor – but also not a unique one.
Why then even bother with courses? Just have a student pay a flat fee & hand them a diploma.