Background <p>Learning health systems (LHSs) aim to integrate continuous learning into routine care, yet their development raises persistent ethical challenges. Questions remain about when and how informed consent should be obtained, how ethical oversight should be organized for learning activities that blur the boundary between care and research, and what system-level conditions are necessary to support ethically sound learning.</p> Objective <p>This integrative review synthesizes ethical recommendations for the design, implementation and evaluation of LHSs, and aims to provide practical guidance and identify gaps in the current literature.</p> Methods <p>A systematic search of PubMed, Web of Science and a search on LinkedIn on 22 October 2025 identified studies offering explicit ethical recommendations related to LHSs. Eligible studies were situated within an existing LHS and provided at least one actionable ethical recommendation. Data were extracted and analysed using Friedman’s LHS functioning cycle of three knowledge-to-action steps: data to knowledge, knowledge to practice and practice to data.</p> Results <p>In total, 44 studies met the inclusion criteria. Ethical guidance was unevenly distributed across Friedman’s cycle. Most studies (<i>n</i> = 32) focused on transforming data into knowledge, addressing data governance, patient autonomy, consent models and ethical prerequisites for artificial intelligence (AI). Far fewer studies (<i>n</i> = 6) examined translating knowledge into practice, where attention centred on the ethical implementation of AI, workflow integration and risk mitigation. The final step, feeding performance back into new data, was represented (<i>n</i> = 6), with limited guidance on accountability, continuous monitoring and equitable interpretation of performance outcomes. Across all stages, informed consent and ethical oversight emerged as dominant themes, though considerable variation existed in how institutions operationalized these concepts.</p> Conclusions <p>Current ethical discourse in LHSs remains focused mainly on transforming data to knowledge. Relatively limited recommendations for ethical implementation of other action steps were identified. In addition, some of the recommendations contradicted each other or offered differing advice on aspects of the ethical implementation of LHS (for instance, on informed consent and ethical review). This imbalance highlights the need for context-sensitive governance models, empirical evaluation of ethical practices in real-world LHSs and regulatory frameworks that reflect the dynamic nature of continuous learning.</p>