Validation of AI and machine learning systems in pharmaceutical quality assurance: A risk-based regulatory approach

ABSTRACTThe increasing adoption of artificial intelligence (AI) and machine learning (ML) technologies within pharmaceutical qualityassurance systems represents a paradigm shift in how qualityrelated decisions are generated, interpreted and acted upon.These technologies offer significant potential to enhance deviation management, predictive quality analytics and proactiverisk mitigation. However, the adaptive and data-driven nature of AI and ML systems presents substantial challenges toconventional computer system validation approaches traditionally applied in Good Manufacturing Practice (GMP)environments. Existing validation models, largely designed for deterministic and static software are often inadequate foraddressing issues such as algorithmic learning, model drift, data dependency, and explainability.This review critically examines the concept of validation as applied to AI and ML systems used in pharmaceutical qualityassurance, with a particular emphasis on adopting a risk-based regulatory approach. The evolving regulatory landscape,including the transition towards computer software assurance and the growing emphasis on patient safety and productquality, is discussed in detail. Key validation considerations such as model lifecycle management, data integrity, biasmitigation, and performance monitoring are explored. Furthermore, the review highlights how risk-based principles can beapplied to tailor validation efforts according to the intended use, criticality, and potential GxP impact of AI-drivenapplications.By integrating regulatory expectations with practical implementation challenges, this review aims to provide a structuredframework for validating AI and ML systems in pharmaceutical quality assurance.