AI doesn't create cheaters: it just makes visible an education system that was already producing them

By Momentum IA · June 30, 2026.

Austin Sarat, professor of Jurisprudence and Political Science at Amherst College, makes a habit of it: at the start of every course he asks his students how many cheated in high school. The answer, year after year, is always the same: most raise their hands, without embarrassment. His analysis, originally published in The Conversation and reproduced in Japan Today, advances an uncomfortable thesis: artificial intelligence did not create the problem of academic dishonesty. It has simply made it more visible, more efficient and harder to ignore.

The figures he cites are forceful. A large national study found in 2018 that 51% of U.S. high school students admitted to having cheated on a test. A 2020 study of 70,000 students raises that figure to 64% on written tests and 58% on plagiarism, and puts at around 95% those who engaged in some form of dishonest conduct —copying on exams, plagiarizing or turning in someone else's homework. At the university level, a 2020 survey of 840 undergraduates found that 32% had cheated on some exam. And a 2024 Harvard Crimson study, with a sample of 850 seniors, found that 47% admitted to having cheated. Cases of misconduct at Ohio State grew by 57% between 2014 and 2018, and that was before generative AI existed as a mass-consumer tool.

The institutional reaction to the rise of AI is, in Sarat's view, a patch over a bone that has been broken for decades. The Wall Street Journal reported in 2025 that many professors are giving up on written assignments and returning to in-person exams. Princeton has just abolished its ban on exam proctoring —which dated back 133 years— citing the proliferation of AI use. Oberlin has amended its honor code to allow professors to supervise tests, which one student described, in a May 2026 op-ed, as a sign that "the school doesn't trust us to learn to be adults with integrity."

**Our reading: the problem is structural, and AI exposes it without causing it**

What Sarat describes is not, at bottom, a story about AI. It is a story about what happens when an education system rewards the numerical outcome —the grade, admission to a selective university— over real learning. Cheating becomes the rational solution for a student who perceives that the goal is to pass the filter, not to understand the subject. Sociologists Sykes and Matza called this "techniques of neutralization": the mind that knows it is doing something wrong builds a narrative that justifies it ("everyone does it," "the teacher doesn't teach well," "it doesn't really count"). It is an ancient cognitive mechanism, perfectly functional without ChatGPT.

What changes with generative AI is the opportunity cost of cheating: it drops to almost zero, the cheating is less detectable to the naked eye, and the scale is different. A dishonesty that once required effort —obtaining someone else's work, paraphrasing without it showing— now takes seconds. But that does not create the disposition; it amplifies one that already existed. The distinction matters because it leads to completely different solutions.

If the diagnosis is "AI makes it possible to cheat," the answer is technological: AI detectors, in-person exams, a return to pen on paper. If the correct diagnosis is "we have trained a generation that has learned to game the system instead of learning," then the problem is one of educational culture, incentives and the construction of intellectual identity. Sarat argues for the latter: treating cheating as a habit that requires a support program spanning four years, not a one-off patch. It is a harder and more expensive proposal, but a more honest one.

There is an angle the article only touches on but that deserves development: AI, in the long run, could actually force a reform the education system had needed for decades. If copying an essay is trivial, essays as we know them cease to be a valid assessment instrument. That forces a rethinking of what an education really measures: the ability to produce text, or to reason, argue and create new knowledge? AI cannot —not yet, and not without a trace— reproduce a genuine Socratic conversation in class, nor the process of a student who builds and defends an idea before a real interlocutor. The institutions that understand this before the rest will have an advantage.

In the short term, however, the transition is real and painful. Universities that try to solve a values problem with in-person proctoring are investing in the wrong direction. Princeton eliminating 133 years of honor culture to introduce exam proctoring is, viewed without condescension, an institutional surrender to a symptom. The root remains intact.

For those working on the design of educational systems or on the institutional adoption of AI, the lesson is clear: new tools inherit old incentives. Deploying AI in classrooms where the implicit goal is still "pass the filter" does not transform education; it automates it more efficiently. The challenge is not technological. It never was.