Decade of AI in Drug Discovery: Hype or Reality?

Abstract

AI promised to reinvent drug discovery: compress timelines, reduce attrition, and unlock previously undruggable drug targets. A decade of substantial capital investment has produced more than 173 clinical programs and landmark structural biology breakthroughs, including the 2024 Nobel Prize–winning AlphaFold platform. Yet as of early 2026, no AI-discovered drug has received regulatory approval. This review analyses the clinical evidence systematically, examining Phase I and Phase II trial outcome data across AI-native biopharma pipelines. The data reveal a bifurcated picture: AI demonstrably outperforms historical industry norms in Phase I (80–90% vs. ~52% success), reflecting genuine superiority in molecular safety engineering and ADMET prediction, while Phase II success (~40%) remains statistically indistinguishable from conventional drug development (~37%). The persistent Phase II failure rate maps onto a single root cause: target hypothesis validity. AI has not yet improved the probability of selecting the correct biological premise. This paper identifies where AI adds value (protein structure prediction, lead optimisation, timeline compression), where it does not (target validation, patient stratification, animal model-to-human translation), and charts the emerging frontier of human-genetics- multi-omics integration and biomedical foundation models as the next phase of AI-enabled progress.