The market has started to say the quiet part out loud. A new PwC survey of more than 1,000 U.S. financial-services executives found that 86 percent consider AI skills training more valuable than an MBA for many new hires. Ninety-one percent are raising compensation for employees with AI capabilities, and a majority will pay a premium for fluency. Elite business schools are responding too, with Wharton and MIT now offering short AI-leadership certificates for a fraction of a degree’s price, even as MBA applications slump and starting salaries drift downward.

For a century, elite college and graduate degrees were the most reliable ladders to prestige and income an ambitious person could climb. They worked because they were costly, hard-to-fake signals: admission filtered for ability, and graduation certified that the ability had been refined and readied for deployment. That bargain is now being attacked from both ends at once. On the demand side, employers are increasingly comfortable discounting the credential in favor of a skill no transcript captures. And on the supply side, the credential is becoming easier to counterfeit from the inside.

For evidence of the second, look to Providence.

A Live Experiment at Brown

Last spring, after nearly two decades teaching Welfare Economics and Social Choice Theory, Brown University economics Professor Roberto Serrano tried something new. A take-home midterm. Students, still rattled by a deadly shooting on campus months earlier, wanted to avoid crowded classrooms. Serrano obliged and made the exam tougher than usual, reasoning that the added flexibility should support deeper analysis.

His midterm averages typically fall between 65 and 80 percent. This time, the class averaged 96. Forty students scored a perfect 100. In a course built around rigor, that number wasn’t a triumph. It was a red flag.

Serrano and his graders ran the exam through ChatGPT. The AI’s answers mirrored many student submissions, at times in substance and in style. One proof was cleanest as a direct argument. However, both ChatGPT and much of the class instead reached for a contorted proof by contradiction—technically valid but not how a human mind naturally attacks the problem.

Convinced dozens of students had leaned on AI, Serrano didn’t void the midterm outright. He gave the class a chance to prove him wrong. If scores on an in-person final matched the midterm distribution, both exams would count. If not, the midterm would be voided and the final reweighted.

The final arrived. Eighteen students dropped the course. Nine stayed enrolled but skipped the exam. Among those who sat for it, the average collapsed to 48.6 percent, the lowest in the course’s history; previous finals had never dipped below 65. Three students scored zero. Nineteen failed the class.

Brown’s response has been procedural: its academic-code committee asked Serrano to file individual complaints against each suspected student, while its Generative AI in Teaching and Learning committee published a report calling AI a “significant disruptor” and urging updates to the academic code.

The machinery is moving. The questions underneath it are not being asked. A signal that is simultaneously less demanded by buyers and cheaper to counterfeit doesn’t collapse overnight. It just quietly stops being worth the tuition. Serrano’s grade sheet and PwC’s survey are the same story told from opposite ends.

What Counts as Cheating When the Machine Is Ubiquitous?

Every technology forces education to renegotiate what counts as your own work. Calculators absorbed arithmetic. Spellcheck absorbed orthography. Search engines absorbed recall. Each time, universities redrew the line and moved the human contribution up a level. You no longer had to compute, but you still had to reason.

Generative AI is different in kind, not just degree. It automates more than a component of thinking. It automates the assembly, the framing, the argument, the proof strategy. It absorbs precisely the thing an exam exists to sample. Honor codes written for plagiarism assume a stable boundary between the mind being tested and the help it isn’t allowed to have. AI dissolves the boundary itself. When the tool sits on every laptop and is rewarded in every internship, students aren’t smuggling something into the exam room. They are doing what the world outside the exam room has already told them is competence.

Look again at what Serrano’s ultimatum produced. Eighteen students dropped the course. Nine stayed enrolled but never showed up. Twenty-seven Ivy League students, offered the chance to demonstrate unaided mastery, declined to try. The easy reading is a confession. The more unsettling reading is a referendum, a cohort quietly rejecting the premise of the test itself.

Because inside that behavior sits an argument, and universities should assume their students hold it even when no one states it aloud. Capability now means what a person can accomplish with the tools they will have—the same shift reshaping careers broadly. On this view, requiring students to suspend AI fluency at the exam-hall door doesn’t reveal the mind; it amputates it. The effort of mastering material the machine already handles starts to look like an arbitrary tax, especially when employers are paying premiums for exactly the skill the honor code treats as a violation. Students who score 96 with AI and 48 without may conclude that the second number measures an increasingly irrelevant version of themselves.

Universities need to take that argument more seriously than they do because their students already live by it. But it fails on two counts and naming them precisely matters more than being scandalized.

First, the offense at Brown was not the presumed usage of AI tools. It was misrepresentation. The students submitted AI’s reasoning as their own, on an exam everyone understood to certify unaided understanding. A student who believes the exam measures the wrong thing has arguments to make and courses to drop; passing off borrowed work as his own is not an epistemological position. It’s deception, and it would be deception in any workplace those students are headed to.

Second, the argument assumes the AI-assisted performance reflected genuinely augmented capability. The final suggests otherwise. Real AI fluency, the kind PwC’s executives are paying for, is the ability to direct the machine, audit its output, catch its errors, and stand behind the result. That requires exactly the domain understanding the 48.6 average showed was missing. A pilot isn’t tested on manual flying because autopilot is cheating; he is tested because he is the fallback when the automation fails. What the take-home midterm measured wasn’t augmented minds. It was the machine, with a student attached.

An Uneasy Equilibrium

So higher education is drifting into an unstable settlement among three forces that don’t yet fit together: a legacy system built to certify unaided minds, honor codes written for a pre-AI world, and a labor market demanding fluency in the very tool the honor codes forbid. Whether universities, let alone individual professors, are equipped to judge students armed with AI is an open question, and the honest answer today is mostly no. Pretending otherwise is how you get a committee asking one professor to file dozens of individual complaints.

Getting to a stabler equilibrium means deciding, explicitly, which thinking must remain the student’s own and which should be examined with the machine in the room. Some coursework should test unaided reasoning through oral defenses and in-person problem solving, because the foundation must exist before it can be augmented. Other coursework should require AI use and grade students on how well they direct, audit, and correct it, making the fluency employers want visible and honest rather than smuggled. And faculty need a way to report suspected mass cheating as a single systemic case, not a paperwork exercise designed for another era.

Serrano put it starkly: “We cannot afford to have a society in which a significant fraction of our best young minds think that cheating is OK. That leads to a declining society, to a failed society.” His grade chart suggests he’s earned the right to say it. But the students’ unspoken rebuttal, that the measuring stick, not the mind, is broken, has earned a hearing too. The institutions that answer both, honestly and in their curricula, will be the ones whose degrees still mean something in ten years. The ones that don’t will keep selling a ladder to a floor the market has already moved.