Artificial intelligence-related patient safety risks in nursing practice: a scoping review

Artificial intelligence (AI) is increasingly being used in nursing practice, including care planning, assessment, documentation, handover, and patient education. These tools may support nurses, but patient safety risks in nursing workflows have not been clearly mapped. To map evidence on AI-related patient safety risks in nursing practice, identify the nursing contexts and use scenarios in which these risks have been reported, and synthesise the mechanisms and safeguards described in the literature. A scoping review. PubMed, Embase, CINAHL, Scopus, Web of Science, PsycINFO, IEEE Xplore, and the Cochrane Library were searched for records on nursing practice, AI, and patient safety. Eligible sources described identifiable AI applications in nursing practice or nursing workflows and reported safety risks, mechanisms, or safeguards. Sources concerning only non-clinical applications, nursing education or research, or technical model development, as well as sources without extractable safety-relevant data, were excluded. Database searches identified 6,295 records up to 28 April 2026, and supplementary searches identified 12 additional records up to 10 May 2026. Eight studies met the eligibility criteria. Most studies evaluated large language models (LLMs) or other generative AI systems in experimental or prototype settings. Four potential patient safety risk categories were identified: inadequate patient assessment and safety-signal recognition; misaligned nursing recommendations and interventions; distortion of nursing documentation and handover information; and decontextualised care with weakened professional judgement. Three mechanism categories were synthesised: unreliable knowledge bases and model performance; misalignment with nursing work systems; and compromised information integrity and nursing representation. Proposed safeguards focused on model optimisation and context-specific validation, human review and workflow control, and governance and competency building. No included study reported direct patient harm attributable to AI use. AI-related patient safety risks in nursing practice may arise through sociotechnical pathways rather than from isolated algorithmic errors alone. However, the evidence remains limited. This review included only eight studies, seven of which focused on large language models or other generative AI systems, and most were conducted in experimental or prototype settings. The findings should therefore not be generalised to all forms of AI or to routine nursing practice. Future studies should evaluate these tools within real-world or simulated nursing workflows and prospectively test safeguards incorporating context-specific validation, nurse-led review, workflow controls, and organisational governance. Registration: OSF ( https://osf.io/uky8g ; DOI: https://doi.org/10.17605/OSF.IO/2G9HZ ).