GAMP 5 AI Gap Report: The 6 Most Common Validation Failures for AI Systems in GxP Environments

Why GAMP 5 AI Validation Is Different

GAMP 5 has been the primary validation methodology for GxP computerised systems since its first edition in 1994. Its category-based risk classification, lifecycle approach, and IQ/OQ/PQ validation framework have guided thousands of system validations across pharmaceutical, biotech and medical device organisations. For conventional software, the framework works well.

AI and machine learning systems challenge several GAMP 5 assumptions. GAMP 5 was designed for deterministic software: given the same inputs, the system produces the same outputs. AI systems — particularly those that learn from operational data — may not meet this condition. A model that was validated against historical training data may behave differently as it processes new operational data over time. The GAMP 5 AI Supplement, published in 2022, extends the GAMP 5 framework to address these characteristics. The six validation gaps in this guide are the most common failures to apply this extended framework correctly.