Abstract Generative artificial intelligence (GenAI) is increasingly used in nursing and health professions education, yet its adoption introduces risks such as hallucinations, misinformation, algorithmic bias, academic misconduct, and erosion of critical reasoning. A structured approach is needed to distinguish system-driven risks from those arising through human, pedagogical, or institutional decisions. An integrative review was conducted following Whittemore and Knafl (2005). Scopus, CINAHL, PubMed, Web of Science, and Embase were searched for literature published from January 2021 to August 2024 using GenAI, education, and risk-related terms. After screening approximately 400 records, 110 full-text articles were reviewed, and about 80 studies were synthesized. GenAI outputs generated from prompts commonly used by students were also examined to identify typical system failure modes. The artificial evil (AE) and moral evil (ME) taxonomy guided data coding and categorization. AE risks included hallucinations, clinical inaccuracies, dataset-driven bias, and output inconsistency. ME risks included over-reliance, academic misconduct, privacy breaches, weak artificial intelligence literacy, and inadequate governance. Interactions between the two domains were evident, with system-level errors amplified by pedagogical, behavioral, or institutional failures. Mitigation strategies centered on technical controls, curriculum redesign, assessment reform, and strengthened governance structures. GenAI risks in health professions education emerge through the interplay of system limitations and human decisions. Integrated strategies that combine technical safeguards with educator and learner capacity-building, clear policies, and redesigned assessment practices are essential to ensure safe, ethical, and educationally sound implementation. Effective integration of GenAI in nursing and health professions education requires coordinated technical, pedagogical, and governance measures to minimize risk and support responsible, evidence-informed use.
Visuranathan et al. (Thu,) studied this question.