Governing Generative AI in Healthcare: A Scoping Review of Liability, Accountability, and Policy Frameworks

1. Introduction and Objectives The rapid advancement and integration of generative artificial intelligence (GenAI) and large language models (LLMs) into healthcare systems present unprecedented clinical, ethical, and legal challenges. While GenAI offers significant potential to enhance clinical decision-making, diagnostics, and patient care workflows, it also introduces complex liability dilemmas due to algorithmic opacity, "black box" decisions, and autonomous error generation. Primary Objective: This scoping review aims to systematically map, characterize, and analyze the existing literature and policy documents regarding legal liability and accountability frameworks for the use of generative AI in healthcare. To fulfill this objective, the review is guided by three specific research questions (RQ): RQ1 (Mapping): What liability and accountability frameworks currently exist for the use of generative AI in healthcare? RQ2 (Characteristics): What are the key characteristics, subjects of liability, and models of accountability identified in the existing literature and policy documents regarding generative AI in healthcare? RQ3 (Gaps & Policy Implications): What are the major policy gaps, challenges, and proposed solutions related to liability and accountability of generative AI in healthcare? 2. PCC Framework (Population, Concept, Context) The scope and eligibility boundaries of this review are structured using the JBI-recommended PCC framework: Population (P): Stakeholders involved in generative AI in healthcare, including patients, clinicians, healthcare institutions, AI developers/providers, regulators, and policymakers. Concept (C): Liability and accountability frameworks, encompassing legal liability, medical malpractice, negligence, duty of care, responsibility allocation, regulatory frameworks, ethical accountability, and governance models related to generative AI. Context (C): Healthcare settings in the generative AI era (clinical practice, public health, medical decision-making, informed consent, diagnosis, treatment, data privacy, etc.), including global and jurisdiction-specific regulatory contexts (e.g., EU, USA, China). 3. Methodology and Literature Search Overview This scoping review protocol follows the methodological framework established by Arksey and O'Malley, advanced by Levac et al., and adheres to the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines. A comprehensive literature search was conducted across four major electronic databases (Scopus, PubMed, ScienceDirect, and Springer Nature) targeting peer-reviewed literature published between January 2023 and December 2026, yielding a total of 1,643 raw records prior to deduplication.