Study Estimates 9 Percent of AI-Using Students Submit Unallowed Work

A Science analysis of 95,513 students at twenty public research universities estimates that about 9 percent of students who use generative AI have submitted AI-generated work they knew might not be allowed. The figure comes from a list experiment designed to reduce underreporting.

Students were randomly assigned to two groups. One group saw three neutral statements about AI use and reported how many applied to them. The second group saw those three statements plus a fourth about submitting AI work as their own when it might not be allowed, and again reported only the total count.

Computer science students report regular AI use at 62 percent, compared with 24 percent in the arts. Cheating rates among users stay narrower, ranging from 5 percent in biology to 17 percent in economics. At the individual level the pattern reverses. Students who use AI daily report a 26 percent cheating rate, while those who use it monthly report 7 percent.

The study also records differences in regular AI use: 33 percent of women compared with 45 percent of men, and 29 percent of underrepresented minority students compared with 39 percent of white and Asian peers. The authors note that a polished final product becomes weaker evidence of unaided student work.

They question the effectiveness of detection tools and AI-resistant exams for measuring the judgment a degree is intended to certify. The survey was conducted in 2024.