The Real Crisis in Higher Ed Isn't AI Cheating—It's How We Test

A junior at a prestigious magnet school recently estimated that about 60% of his peers use AI to gain an unfair advantage. In an informal X survey, many professors and teachers put the number at at least 40% of students submitting AI-generated assignments. These figures, reported by The Washington Post in September 2024, point to a crisis that has only deepened since. But the real story is not about cheating—it’s about how universities still test students in ways that AI can easily replace, and what that means for the future of learning.

The cheating numbers are stark. A Sept. 30, 2024 Washington Post opinion piece framed the issue as an existential threat to colleges, arguing that the difficulty of assessing student learning in the AI era is the core problem. Since then, evidence has piled up at elite schools. In July 2026, the Post reported that Brown University professor Roberto Serrano suspected AI cheating after his advanced mathematical economics midterm average jumped to 96, compared with 60s to 80s in previous years. Nearly half the class scored 100—a statistical anomaly that forced the professor to investigate.

A Feb. 24, 2026 Post story, citing Pew data, found that nearly 60% of teens believed students at their schools use AI to cheat “somewhat often,” and about 12% said they used AI for emotional support or advice. The scale is undeniable.

The response from universities has been to double down on AI-detection tools. But those tools have documented false positives and are not considered reliable enough by some institutions. A University of Maryland analysis, reported by The New York Times, found that AI detectors falsely flagged human-written text as AI-generated about 6.8% of the time on average. Some schools have already disabled Turnitin’s AI-detection feature over reliability concerns. The risk of punishing innocent students is real, and it erodes trust in academic integrity systems.

Yet the deeper problem remains: the tests themselves are broken. If a chatbot can ace a midterm, what does that exam actually measure? The Post’s columnist argued that the issue exposes how hard it is to assess student learning in the AI era. This is not a technology problem—it’s a pedagogical one. The same tools that students use to cheat can also be used to create more meaningful, project-based assessments that require critical thinking and original analysis. But most universities are still stuck in a model of high-stakes, recall-based testing that AI can replicate in seconds.

For employers, this is a warning signal. Hiring managers already struggle to interpret transcripts and GPAs. If a significant portion of coursework is AI-generated, the signal becomes noise. Companies that rely on university credentials as a proxy for skills will need to rethink their hiring pipelines. Some are already moving toward skills-based assessments, coding challenges, and portfolio reviews—methods that are harder to fake.

For edtech companies, the opportunity is clear: build assessment tools that are AI-resistant by design. Oral exams, which some instructors are bringing back to make cheating harder, are one example. But scaling them is expensive. Startups that can create secure, proctored environments or design assessments that require real-time interaction and original thought will find a ready market.

The most interesting development is the return of oral exams. The Washington Post reported that some instructors are bringing back oral exams to make cheating with AI harder. This is a throwback to medieval universities, but it may be the only reliable way to test deep understanding in a world where written assignments are easily automated. Expect more universities to experiment with hybrid models: written work for drafting, oral defense for verification.

Also watch for regulatory moves. As AI cheating becomes a public scandal, lawmakers may pressure accrediting bodies to set minimum standards for assessment integrity. And universities that fail to adapt may see enrollment drops as students question the value of a credential that no longer proves anything.

For now, the numbers speak for themselves: 60% of peers using AI, 96% class averages, 6.8% false positive rates. The system is broken. The question is whether universities will fix the tests or just keep chasing the cheaters.