PATIENT SAFETY IN THE ERA OF ARTIFICIAL INTELLIGENCE: OPPORTUNITIES, ETHICAL CHALLENGES AND FUTURE PERSPECTIVES

Abstract

Objective: to critically synthesise the best scientific evidence on the application of artificial intelligence (AI) to patient safety, analysing its potential to prevent adverse events, the ethical risks and challenges arising from its incorporation, and future perspectives for clinical practice and health systems governance. Method: integrative review conducted according to Whittemore and Knafl and reported in accordance with PRISMA 2020, with searches across eight databases, DeCS/MeSH descriptors combined through Boolean operators, covering 2020–2026, selection by two independent reviewers, appraisal of the level of evidence (Melnyk and Fineout-Overholt) and risk-of-bias assessment with a prediction-model-specific tool (PROBAST+AI), complemented by the CONSORT-AI and TRIPOD+AI reporting lenses. Results: twenty studies (2020–2026) were assembled, predominantly prediction-model development/validation studies and scoping reviews, addressing sepsis and deterioration prediction, medication safety, pressure injury and fall prevention, NLP-based adverse-event surveillance and decision support. Benefits were conditional on implementation quality, with recurrent limitations: internal-only validation, surrogate outcomes, false alarms, automation bias and algorithmic inequities. Final considerations: AI offers concrete opportunities to prevent adverse events, but its benefits depend on external validation, meaningful human oversight, governance and continuous monitoring; nursing and the multiprofessional team play a central role as a safeguard against automation bias.