On Campus, More AI Use Means More Cheating. Across Majors, It Means Less

A sweeping new study of nearly 100,000 college students has uncovered a paradoxical relationship between AI use and cheating: across entire majors, more AI adoption correlates with less cheating, yet students who lean heavily on the technology are far more likely to cross ethical lines. Published in Science on May 25, 2026, the analysis of 95,513 students finds that while AI use barely predicts cheating at the aggregate level, heavy individual users cheat at significantly higher rates. The findings challenge simplistic narratives and force educators to rethink grading in the age of generative AI.

Since the release of ChatGPT in late 2022, colleges have scrambled to develop policies around AI. Early panic about widespread cheating gave way to a more nuanced debate about how to integrate AI into learning. The new study provides the largest empirical test yet of the link between AI usage and academic dishonesty. Researchers surveyed students across multiple majors and institutions, measuring both self-reported AI use and instances of cheating, while controlling for factors like GPA, course difficulty, and demographic variables.

Key results show that at the departmental level, a 10% increase in AI adoption was associated with a 3% decline in cheating incidents. But at the individual level, students in the top 10% of AI usage were 40% more likely to report cheating than those in the bottom 10%. This decoupling between macro and micro trends suggests that AI is not inherently corrupting. Majors where AI is integrated as a tool—such as computer science or engineering—may foster a culture where AI aids learning rather than shortcuts it. Conversely, students who use AI to generate entire assignments may be those already predisposed to cheat.

The implications for grading are profound. The study’s authors caution against blanket bans on AI, advocating instead for “AI literacy” and adjusted assessment designs. Traditional exams that test memorization may become obsolete; open-book, collaborative, and project-based evaluations may better capture genuine learning. Universities that embrace AI as a teaching ally rather than an enemy may see improved outcomes without sacrificing integrity. For now, the message is clear: context is everything. The research, reported by Forbes, calls for more nuanced policies that account for both aggregate trends and individual behavior.