Across California’s public universities, students say anti-AI rules have jumped from fine print in the syllabus to something that feels a lot like surveillance: mirrors propped behind laptops, remote proctors monitoring every corner of a desk and oral exams where students have to sit with their arms crossed so they physically cannot type. The goal is to block ChatGPT-style shortcuts, but many students say they are being flagged simply for clean prose, consistent revision or a well-edited paragraph. The result is a classroom climate where trust, once the quiet backbone of instruction, is starting to come apart.
How campuses are policing AI
These stories are not campus folklore. As reported by the Los Angeles Times, UCLA students say they have been told to “procure a mirror large enough to fully reflect your entire desk-area” and to take oral exams with their hands behind their heads. Attorneys who represent students report that AI-related accusations now make up a large share of their education caseloads. The Times also found that faculty practices can shift dramatically from one class to the next, leaving students unsure when AI help is allowed and when it is grounds for discipline. Those gaps have led to contested hearings and, in some cases, lawyers stepping in to defend students who insist they wrote their own work.
What the research shows
A new study helps clarify the scale of AI use behind all this tension. A major paper led by Igor Chirikov at UC Berkeley surveyed more than 95,000 undergraduates and found that roughly two-thirds had used generative AI for classwork, about one-third used it regularly and nearly 9% said they had used it to cheat in ways that violated course rules. The authors say both use and misuse look very different across disciplines and they recommend “discipline-specific assessment reform” instead of blanket bans. UC Berkeley’s Center for Studies in Higher Education has summarized the work in detail, and UC Berkeley’s CSHE lays out the study and its implications.
Why detectors aren’t a smoking gun
Many instructors are turning to AI-detection tools to sort through suspected cases, but those systems come with significant caveats. Turnitin, one of the dominant companies in the space, says its document-level false-positive rates are very low when a paper contains substantial AI-generated text and stresses that AI scores should trigger human review, not serve as final proof in disciplinary cases. That distinction matters because even a small error rate, scaled across large lecture courses and millions of submissions, can still generate plenty of mistaken flags and painful investigations. Turnitin outlines those limits and the use guidelines it recommends.
Integrity experts say institutions aren’t ready
Academic-integrity experts say many campuses are still scrambling to catch up. Tricia Bertram Gallant, director of UC San Diego’s Academic Integrity Office, has argued that colleges need clearer systemwide guidance and redesigned assessments so instructors can better judge authorship while also teaching students how to use AI ethically. UC San Diego’s Academic Integrity Office has published resources and training materials to help faculty and staff navigate generative AI in coursework. UC San Diego’s AIO documents that work and its recommendations.
Rethinking assignments, not just policing
Students are adjusting in real time. Many now hang on to drafts, save timestamped files and keep detailed revision logs so they can show how an essay evolved, and some turn to attorneys when allegations escalate. Chirikov and his coauthors argue that because GenAI adoption varies so much by discipline, colleges should rethink how they assess learning rather than relying on one universal enforcement playbook. That could mean more in-person, performance-based tasks in some courses and AI-integrated assignments that explicitly require attribution in others. Those recommendations nudge institutions away from one-size-fits-all detection regimes and toward more targeted instructional changes. UC Berkeley’s CSHE describes those proposed shifts.
What to watch next
Administrators are already in motion. The University of California system and several campuses have convened work groups and issued guidance as they rewrite academic-integrity rules for the GenAI era. For students, that means watching syllabi and class-by-class AI policies closely and keeping clear records of their work. For now, experts say the strongest protections are transparent rules, human review of any AI flags and assignment designs that actually test what students know rather than how cleverly they can write a prompt. The UC Senate outlines systemwide recommendations administrators are weighing.
