Abstract

The pharmaceutical data gap, which includes the systematic under-representation of female patients in research investment, clinical trial populations and adverse event reporting, does not resolve at any stage of the drug development pipeline. It compounds. This article traces the mechanism by which structural bias enters through early-stage research funding decisions, persists through absorption, distribution, metabolism, excretion and toxicity (ADMET) modelling and clinical trial design, and reaches its most consequential expression in AI-augmented post-marketing drug safety surveillance systems trained on decades of skewed spontaneous reporting data.

Drawing on the 2025 CIOMS Working Group XIV report, the 2024 European Medicines Agency (EMA) reflection paper on artificial intelligence (AI), the 2025 US Food and Drug Administration (FDA) draft guidance on AI in drug development, the EU AI Act and the 2026 World Health Organization (WHO) European region findings, this article argues that four converging regulatory frameworks create both an obligation and a defined compliance horizon for action. The EU AI Act’s August 2026 compliance deadline for high-risk AI systems has now passed, marking a critical inflection point for marketing authorisation holders (MAHs) operating in Europe. The article concludes with targeted recommendations for MAHs seeking to address inherited bias in deployed pharmacovigilance AI, with the patient as the ultimate subject of governance failure.