Artificial Intelligence in Intensive Care: From Clinical Applications to Ethical and Medico-Legal Challenges

From a medico-legal perspective, AI functions as a clinical decision support tool without legal autonomy. Responsibility for AI-assisted clinical decisions remains subject to the applicable legal and regulatory framework and depends on the degree of human oversight. The use of artificial intelligence (AI) is rapidly transforming intensive care medicine, offering advanced tools for data processing, prediction, and clinical decision support. This narrative review explores current applications of AI in the intensive care unit (ICU), along with associated ethical, medico-legal, and regulatory challenges. AI systems, particularly those based on machine learning and deep learning, have demonstrated promising predictive performance across several intensive care applications. Systematic reviews of machine learning models for ICU prediction tasks have reported AUROC values generally ranging from approximately 0.80 to over 0.90 for early sepsis prediction and ARDS risk stratification, although performance decreases in external validation cohorts. AI has also shown encouraging performance in predicting weaning success and supporting mechanical ventilation management, while evidence of improved patient-centered clinical outcomes remains limited A narrative literature search was conducted in PubMed and Scopus for studies published between January 2015 and December 2025. Additionally, AI contributes to continuous patient monitoring, reduction in false alarms, and improved workflow efficiency. Despite these advances, most models are derived from retrospective and single-center datasets, limiting generalizability and real-world applicability. Moreover, only a limited number of AI systems have undergone prospective clinical evaluation or demonstrated improvements in patient-centered outcomes following real-world implementation. A key limitation of current AI systems is the “black box” nature of many algorithms, which reduces interpretability and may undermine clinician trust. Furthermore, algorithmic bias, data quality issues, and limited external validation raise concerns regarding safety, equity, and reliability. Importantly, improved predictive accuracy does not necessarily translate into better clinical outcomes, highlighting the need for prospective validation studies. From a medico-legal perspective, AI functions as a clinical decision support tool without legal autonomy. Responsibility for AI-assisted clinical decisions remains subject to the applicable legal and regulatory framework and depends on the degree of human oversight, the intended use of the AI system, and the relevant jurisdiction. The emergence of regulatory frameworks, such as the European Union Artificial Intelligence Act and the General Data Protection Regulation (GDPR), reflects the increasing regulation of healthcare AI according to the intended purpose, level of autonomy, and potential impact of each system, emphasizing requirements for transparency, human oversight, and data protection.