The take-home exam is over
A professor at Brown denounced mass AI cheating on a take-home exam. The outrage landed on the students. The design problem deserves more attention.
A professor at Brown University went public last week describing what they called mass AI fraud on a take-home exam. By their account, almost the entire class had used AI tools to complete it. The professor was furious. The students, apparently, were not.
I have a hard time sharing the outrage.
The exam was closed-book, unsupervised, completed at home over the internet. In 2026. The expectation was that students wouldn’t use AI because they had signed an honor code. Leaving a bicycle unlocked in a crowded station can work. This is roughly that kind of optimism.
When not cheating becomes irrational
Consider the position of any individual student. They know the exam is ungoverned. They know AI is available to every classmate. If they believe even a few peers might be using it, choosing not to use it just costs them their grade positioning. The stakes are real: grades shape graduate school applications, job offers, fellowships.
This is a prisoner’s dilemma. When defection is private, costless, and potentially universal, you get mass defection. The exam design created exactly this structure. Framing it as a failure of student ethics doesn’t solve anything structural.
What the test was actually measuring
Closed-book exams were always a proxy. The nominal goal was to check whether students had absorbed material, but the real target was something harder to look up: synthesis, judgment, application to novel problems. The format kept evolving to stay ahead of what students could look up.
AI collapsed that proxy in one move, and the exam design didn’t follow.
A model can recall course material better than any student. It structures arguments fluently. What it can’t show is that the student understands what it produced (whether they can defend it, extend it, spot where it goes wrong). That’s the evidence of real learning, and no take-home exam can capture it anymore.
What actually works now
The solution is redesigning the assessment. Honor codes won’t hold against AI that is free, instant, and invisible.
Oral exams, in-person problem-solving under observation, project defenses where students answer follow-up questions on their own work. These are harder to run but they test judgment rather than retrieval. A student who used AI to draft an answer but can walk you through the reasoning learned something. One who submitted AI output they can’t explain didn’t, and an oral exam makes that apparent in five minutes.
The scaling argument is real: a professor handling 200 students can’t run individual oral exams. That constraint deserves a real answer. The honor code has already been tested and found wanting.
Across my work, when I review someone’s output, I care less about whether they used a tool than whether they can stand behind it. If you show me analysis and can’t answer a basic follow-up, the analysis means nothing regardless of how it was produced. That’s the standard that matters in any real professional context. Universities could adopt it instead of defending an exam format that was already struggling before AI arrived.
The Brown professor won’t be the last to encounter this. Each time, the story will frame it as a crisis of student ethics. The slower, harder question (whether closed-book take-home exams make sense when AI is this accessible) is worth getting to sooner.