University administrators are wrestling with how to design AI policies when actual student use is hard to measure. A University of Chicago study based on an anonymous survey of 338 undergraduates finds a sizable gap between self-reported AI use and students’ perceptions of peers’ use—students say they use AI at 60% but believe the average student does so at 90%. Researchers behind the study—including computer scientist Alex Kale—warn that policy built on uncertain assumptions risks being miscalibrated. They point to social desirability bias as one explanation, where students may underreport usage to avoid appearing dishonest or unable to complete coursework independently. The finding comes as colleges scramble to regulate classroom AI. It underscores the need for institutions to combine policy with measurement approaches and assessment redesigns that evaluate authentic understanding rather than policing tool use alone.