Assessment validity attrition under deep AI involvement in medical education

Supporting files for a scoping review (PRISMA-ScR reported) of 194 studies examining how generative AI involvement degrades the validity of inferences drawn from learner assessment in medical and health-professions education. Four sources (PubMed/MEDLINE, ERIC, OpenAlex, Crossref) were searched on 20 October 2026 with a reproducible three-block query. Of 949 retrieved records, 755 were excluded with mutually exclusive reasons and 194 were included. Every included study was coded independently by two reviewers; Cohen's kappa was 0.515 / 0.739 / 0.144 / 0.732 across the four dimensions (mean 0.533), with 71/194 (36.6%) exact agreement and an Overall Jaccard Index of 0.563. All records are bibliographic metadata for published literature. The dataset contains no personal data, no patient data and no identifiable information about any individual. The full pipeline (search strings, deduplication, screening decisions, dual coding, adjudication) is documented in the files provided so that the review can be re-executed. Citation: Tan, W., Yang, Y., & Wang, X. (2026). Assessment validity attrition under deep AI involvement in medical education: A scoping review and conceptual framework [Data set]. Open Science Framework. https://doi.org/10.17605/OSF.IO/VFGEQ License: CC0 1.0 Universal.