Folks, today's AI faceplant story deserves the first chapter in the 2026 Compendium of AI Education Disasters.

At Ivy League Brown University, economics professor and renowned game theory economist Roberto Serrano recently nearly flipped his desk over one course's exam results. What course? An upper-division economics class. What exam format? Take-home closed-book exam.

You'd think for a take-home closed-book exam, students would be on their honor, right? This is Brown, an Ivy, America's elite among elites -- academic integrity should be in their bones, right?

So what happened? The class average was 96%.

What does 96% mean? The historical average for this course is 60-70%. Dozens of students in one class, nearly everyone approaching perfect scores. Professor Serrano has taught for years and seen everything, but this score distribution -- he used one phrase: "statistically impossible."

Let that sink in. A class that normally averages 60s-70s suddenly has everyone scoring 94-96? Did the entire class simultaneously undergo genetic mutation and collective enlightenment? Did the gods of academia descend en masse on exam day?

Professor Serrano's judgment was direct: mass AI cheating. Take-home exam, no proctoring; students fed questions straight to ChatGPT or Claude, then copied back the answers. That 96% isn't their ability -- it's AI's ability.

The AI Cheating Arms Race

This story blew up on Hacker News: 393 upvotes, 517 comments, hitting the top three on the front page.

The comments section was pure brutal honesty. A professor weighed in: "I gave up take-home exams last year. Online exam averages are 15-20 points higher than in-person, and that AI-specific 'correct but hollow' tone in answers is instantly recognizable." A student admitted: "It's not that we want to cheat; if everyone else uses AI and you don't, you get curved down." Another quipped: "The irony is professors use AI detection tools to catch cheating, and students use better AI to bypass detection -- it's literally an arms race, and the defense is always one step behind."

Folks, this is the awkward reality of education in the AI era: AI cheating detection technology will never catch up to AI cheating technology. The road is one foot high, the demon is ten feet high, and the demon updates faster.

What's even trickier: some students don't just copy AI answers directly -- they have AI generate an answer framework, then rewrite it in their own words -- making it undetectable even by AI detectors. That's the situation Professor Serrano faces: answers look human-written, but the score distribution is all wrong.

Education's Ultimate Question: What Are We Actually Testing?

Brown's story is superficially about cheating, but the deeper issue is the education assessment system.

When AI can answer exam questions faster and more accurately than the vast majority of students, what are we actually testing? Memorization? Problem-solving speed? Or the areas where AI falls short -- critical thinking, creative thought, complex problem decomposition, cross-domain synthesis and judgment?

Apple passed the costs of AI data centers hoarding memory straight onto MacBook price tags -- someone always pays AI's bills; the only question is who. Education is the same: if exams only test questions AI can answer, the exams themselves lose meaning.

Of course, this isn't making excuses for cheating. Cheating is cheating; academic integrity is academic integrity. But universities also need to reflect: in an era where AI is always available, what assessment methods can genuinely test student ability? Closed-book exams back in exam halls? Open-book exams requiring oral defense? Project-based assessment replacing exams? These are all directions being discussed, but there's no standard answer yet.