Frequent Use, Persistent Concerns, and Limited Institutional Support: A Multi-Institutional Survey of Generative Artificial Intelligence Among Health Professions Educators and Trainees

Background: Generative artificial intelligence (GenAI) is now part of routine academic work in health professions education, often outpacing institutional policy, infrastructure, and training. We describe how educators and trainees use GenAI, their concerns, and the institutional support they perceive, and examine whether prior AI training and frequency of use are associated with these perceptions. Methods: We conducted a cross-sectional online survey (August–November 2025) of faculty, clinicians, staff, and trainees involved in health professions education across the United States, recruited through professional networks. Ten items on a 0–5 agreement scale captured attitudes, concerns, trust in AI accuracy, and perceived institutional resources. We used descriptive statistics, Spearman correlations, Mann–Whitney U tests with Holm correction, and multivariable ordinal logistic regression with sensitivity analyses. Results: Of 211 responses, 183 respondents with documented consent were analyzed (76.5% faculty; 48 named institutions). Nearly all had used GenAI (98.9%), and 78.1% used it at least weekly, most often for writing or editing (77.6%). Half or more agreed or strongly agreed that they were concerned about academic integrity (58.5%), data privacy (56.3%), and professional accountability (50.8%), yet only 14.2% agreed or strongly agreed that they trust the accuracy of AI-generated content. Only 24.0% agreed or strongly agreed that their institution provides the resources they need (M = 2.51). Among respondents with elevated concern, 73.8% still used GenAI weekly or more. Any AI training (39.9%, mostly informal workshops) was associated with greater comfort and perceived institutional resources, but not significantly with trust or any concern. More frequent use was independently associated with greater trust (OR 4.93, 95% CI 2.46–9.86) and fewer ethical concerns (OR 0.31, 95% CI 0.16–0.61), and both associations held in sensitivity analyses. Privacy concern was not associated with use frequency. Conclusions: GenAI has become part of routine work for most health professions educators, yet it coexists with substantial concerns about privacy, integrity, and accountability. Training was associated with comfort but not with lower concern, suggesting that faculty development alone is unlikely to be sufficient; institutions may need to pair it with the governance, data protection, and approved tools that individual training cannot provide.