The Race to Keep Cheating With AI
The academic-dishonesty crisis is well documented. When the Harvard Crimson surveyed undergraduates this spring, nearly 40 percent copped to using AI “in ways their instructor may view as inappropriate or against class policy.” Now, schools are scrambling to catch up. Princeton and Oberlin — institutions that have long prided themselves on faith in undergraduates’ morality — are ending the tradition of unsupervised exams. The University of Chicago’s social-sciences division is banning all technology from classrooms in required courses.
But as some faculty move to restrict devices, many administrators are betting on something else: better tech. When California State University system schools upgraded their contracts with Turnitin, a plagiarism-detection giant, campuses paid an additional fee for a tool that scans essays for AI. Columbia University’s Teachers College recommends educators use Draftback, a Chrome extension that lets professors replay a Google doc’s entire revision history. Other institutions have GPTZero, a program that promises to find AI-powered plagiarism with “99 percent accuracy.”
Despite all this, as the fourth academic year of AI college begins, students keep finding innovative ways to skirt the rules. Just a year before Jared’s friend pitted an LLM against WebAssign, agentic coding tools couldn’t yet reverse-engineer the weaknesses in professors’ grading platforms, much less build bespoke software to exploit them. But in the spring of 2025, a new class of AI tools became available to the public. Unlike mere chatbots, newer models can plan multistep tasks, call on external tools like web search and code execution, and carry out complex assignments — from building slide decks to writing and running code — with minimal human supervision.
