The integration of artificial intelligence (AI) and machine learning (ML) into pharmaceutical, biotechnology, and medical device industries presents unprecedented opportunities for enhancing product quality, patient safety, and operational efficiency. However, the inherent characteristics of these technologies, including their adaptive nature, complexity, and potential opacity, pose significant challenges to traditional validation paradigms established under Good Practice (GxP) regulations. While several recent reviews have mapped the emerging regulatory landscape, the literature has stopped short of critically comparing how different jurisdictions approach AI governance, of engaging with the validation implications of generative AI, or of translating high-level principles into an operational framework that practitioners can apply. This review addresses those three gaps. First, we synthesize current regulatory frameworks from the United States Food and Drug Administration (FDA), European Medicines Agency (EMA), and the European Union Artificial Intelligence Act (EU AI Act), including the newly released EU GMP Annex 22 and the FDA’s January 2025 draft guidance on AI for drug and biological product regulatory decision-making. Second, we perform a structured comparative analysis that surfaces tensions between the FDA’s guidance-based, risk-based flexibility and the EU’s more prescriptive, legislatively anchored approach, and we trace the ethical governance implications of those differences. Third, we develop the Integrated Validation, Ethics, and Lifecycle (IVEL) framework, a seven-stage actionable workflow that unifies regulatory, technical, and ethical considerations for AI/ML deployment in GxP settings. The IVEL framework is presented as a proposed integrative workflow that organizes existing regulatory and quality requirements into a sequenced structure, not as an empirically validated standard; its practical application is expected to vary with organization size, AI maturity, and regulatory jurisdiction. Additionally, this review dedicates a substantive discussion to generative AI and large language models, examining hallucination risk, prompt sensitivity, evaluation metrics, and the narrow scope EU GMP Annex 22 carves out for such systems. We illustrate the framework through four case studies, two of which are expanded to explicitly analyze failure modes, over-reliance dynamics, hidden bias, and limits of generalizability. By synthesizing current regulatory guidance with a comparative critique, a practical validation framework, and an engagement with emerging AI paradigms, this article provides both a conceptual contribution and a practitioner-facing roadmap for organizations seeking to implement AI/ML technologies while maintaining compliance with GxP requirements and upholding patient-centric ethical standards.
Patel et al. (Tue,) studied this question.