How Instructors Regulate AI in College: Evidence from 31,000 Course Syllabi

The way instructorssupport or restrictstudent use of generative artificial intelligence (AI) tools shapeswhich skills students developin collegeand their  preparation for AIaugmented jobs. This paper introduces a taskbased approach toAI and skill formationtostudyhow instructors regulate student use of AI, predictingthat regulation depends on course task composition and the strength of AI capabilities in those tasks. I test these predictions using 31,000 course syllabi representing the full universe of courses at a large public research university from 2021 to 2025,providing novellargescale longitudinal evidence on instructor responses to AI. I use computational methods to extract required course tasks and AI policies from each syllabus, tracking the same courses and instructorsover time. Idocumentthreemain results. First, instructors are warming toward AI over time, shifting from initial restrictive policies toward greater permissiveness, differentiating policiesbytasktype,andincreasinglyintroducingnew AIbasedtasks,withsubstantial disciplinary variation. Second, course task composition in 202122 (preChatGPT) predicts AI regulation in 2025: courses requiring more practice in tasks with strong AI capabilities are more likely to regulate students’ use of AI. Third, courses requiring more tasks in 202122 are more likely to differentiate permissions by task type in 2025, consistent with diverse task requirements creating both risks and opportunities for AI affecting task practice. Taken together, these findings indicate a move toward nuanced regulation aligned with tasklevel concerns about skill formation.