On Sunday, Roberto Serrano — the Kravis University Professor of Economics at Brown — went public with what he calls “overwhelming evidence” that at least 50 of his students used AI to cheat on a March midterm in ECON 1170, an advanced mathematical economics course. The median score was 98. Forty out of 86 students scored a perfect 100. Serrano, who published his findings through a university process that concluded last week, described it as the largest known academic fraud case in Ivy League history.

The conventional reaction has been swift and predictable: outrage at the students, lamentations about the death of academic integrity, and a fresh round of hand-wringing about whether ChatGPT is turning a generation of elite undergraduates into high-performing plagiarists. By Monday morning, the story had climbed to the top of Hacker News, where 551 comments debated detection tools, honor codes, and whether the students should be expelled.

That conversation is easy to have. It is also, almost entirely, a conversation about the wrong thing.

The Exam That Shouldn’t Have Existed

Buried in the coverage is a detail that should stop anyone who has ever taught a college course cold: this was a take-home, closed-book exam. Not an in-person blue-book final administered in a proctored lecture hall. Not an oral examination. Not a timed, locked-down computer-lab assessment. A take-home exam — the kind where a student receives a PDF, retreats to their dorm room, and is trusted to return a few days later having used nothing but their own brain and a calculator.

Serrano, to his credit, had a reason for the format. According to the Brown Daily Herald, he normally holds in-person exams but switched to take-home this semester to alleviate pressure on students after the December 13 shooting near campus. That is a humane, defensible decision made under real duress. But it also reveals something uncomfortable: the take-home exam wasn’t a pandemic-era improvisation that should have died in 2022. It is still, in 2026, a live option in the assessment toolkit at one of the eight most selective universities in the country.

And that is the real scandal — not that 50 students cheated, but that a tenured professor at an Ivy League institution handed out an unproctored, asynchronous, take-home exam and then seemed surprised when the grade distribution came back as a vertical line at 100.

The Assessment Problem That Nobody Wants to Name

The uncomfortable question the Serrano case raises is not “how do we catch AI cheaters?” It’s “why are we still designing assessments that can be beaten by a $20 chatbot?”

An advanced mathematical economics course at Brown is, presumably, a rigorous thing. The students are bright. The material is dense. But the exam itself — a set of written problems, submitted remotely, graded on a curve — is a format that has been vulnerable to off-the-shelf AI for at least two years. GPT-4 was released in March 2023. Claude and Gemini have only gotten better at formal mathematical reasoning since. If your assessment can be aced by a student who doesn’t understand the material but does know how to copy-paste a problem set, the assessment is doing none of the work it was designed to do.

This is not a new observation, but it is one that the higher education establishment has been remarkably successful at ignoring. The post-pandemic consensus has been to invest heavily in detection — AI checkers, plagiarism software, honor-code tribunals — while leaving the underlying assessment architecture largely untouched. The logic is, in its own way, a kind of institutional magical thinking: if we can just catch the cheaters fast enough and punish them harshly enough, the incentive structure will hold. But detection is a cat-and-mouse game that scales poorly, and a 50-student fraud case at a single university suggests the mice are winning.

“It’s not that the students are especially devious,” said one teaching-track economist at a large public university, speaking over coffee between sessions at a conference. “It’s that we’ve given them an assessment that is, on its face, a test of their ability to use AI well.baked into a take-home format. Of course they’re going to use it.”

What In-Person Actually Means

Serrano’s own response is, in some ways, the most telling part of the story. He has decided to return to in-person exams for all his courses. The Brown Daily Herald reported that shift in April, before the fraud findings were formally concluded. It is a sensible, obvious move. But it also raises a question that the university — and every university — will have to answer in the next few years: if the solution to AI cheating is to make every high-stakes assessment in-person, proctored, and analog, what exactly was the value proposition of the take-home infrastructure that universities spent a decade building?

The answer is that it was never about pedagogy. It was about convenience, and about cost. Take-home exams, online submissions, and asynchronous assessment formats reduce the demand on physical classroom space, simplify scheduling, and lower the logistical overhead of running large courses. They are, in the language of university administration, “efficient.” The problem is that they are efficient at producing grades that no longer measure anything.

A university that wants to maintain the integrity of its degree — and Brown’s degree is, on the open market, worth a great deal — will have to accept that some of that efficiency is going to have to be sacrificed. Proctored, in-person, closed-device exams are more expensive to administer, harder to schedule, and less flexible for students. They are also, at this point, the only assessment format that produces a signal you can trust.

The Real Debate

Serrano is right that the moment calls for an “in-depth debate.” But the debate should not be about whether the students who cheated are morally culpable — they are, and the university’s disciplinary process will sort that out. It should be about whether the institutions that designed the assessments are institutionally negligent.

A university that hands out a take-home exam in 2026 is not testing its students’ knowledge of mathematical economics. It is testing their willingness to follow an honor code that the technology environment has rendered nearly unenforceable. Some students will follow it, because they take the norms of academic life seriously or because they fear the consequences of getting caught. Many will not. The ones who do not are not, in any meaningful sense, failing the course. They are passing a different test — one that the university did not intend to give, but gave regardless.

If the 50 students at Brown had flunked that exam, the story would be about a professor whose assessment was too hard. Because they aced it, the story is about a professor whose assessment was too easy to cheat. The common thread is that the assessment itself was the problem, and no amount of detection or discipline will fix it. The only fix is to stop handing out exams that can be taken by anyone with a laptop and an internet connection, and to start handing out exams that can only be taken by the students who actually show up.

That is not a technological solution. It is an institutional one, and it will be more expensive and less convenient than the alternative. But the alternative, as Serrano just discovered, is a transcript full of grades that mean nothing at all.

Sources