Answers are grounded in this paper and cite their sources.
Introduction Generative artificial intelligence (GenAI) is increasingly used in nursing education, but studies rarely evaluate task delegation explicitly. This review mapped evidence on educational tasks, contexts of use, and safeguards to inform boundaries for using large language models (LLMs) and delegating educational tasks. Methods This scoping review with critical evidence synthesis used JBI guidance for evidence mapping and PRISMA-ScR for reporting. MEDLINE, Scopus, and Web of Science were searched through 7 August 2026 for studies involving direct interaction between prelicensure nursing students and LLM-based systems. The review team extracted outcomes, risks, safeguards, and human oversight. The team proposed delegation statuses by considering the consequences of error, error detectability under specified safeguards, the need for human judgment, confidence in the evidence, and oversight feasibility. Results Of the 440 records screened, 46 studies comprising 47 educational application profiles were included. Of the 146 outcome records, 115 were favorable, 13 were mixed, 10 showed no difference, 6 were unfavorable, and 2 were unclear. Twelve studies (26.1%) reported at least one objectively verified error or event or an adverse educational outcome. Among the 118 safeguards, 100 (84.7%) were proposed or implemented without evaluation beyond implementation, five (4.2%) were evaluated for feasibility or acceptability, and 13 (11.0%) were evaluated for risk reduction. The proposed matrix comprises 16 task categories: eight conditionally delegable, five restricted to a supportive role, two unsuitable for automated delegation, and one with insufficient evidence. Conclusion The matrix offers a preliminary framework for task-specific decisions. Bounded knowledge support and low-risk formative feedback may be conditionally delegable with appropriate safeguards and oversight. AI should support clinical reasoning and care planning, whereas final grading, competence decisions, and ethical judgment should remain human responsibilities. The matrix requires validation and updating as direct evidence, particularly on safeguard effectiveness, becomes available. Clinical trial number Not applicable.
No takes yet. Share an insight, caveat, or question.
Zare et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: