Cytiva Says Drug AI Models Fail On 90% Trial Rates Due To Incomplete Data

The Common Question

Why is AI not improving drug discovery success rates yet?

The Lesson in This Case

Drug development already costs $1 billion to $2.5 billion per candidate and still fails more than 90% of the time – and AI's ability to accelerate hit identification is now outpacing the lab infrastructure and data quality required to act on those hits. The bottleneck is not the algorithm; it is the fragmented, bias-riddled, and often siloed data that AI models are trained and validated against, including a near-total absence of negative experimental results that would make predictions more reliable. Senior leaders funding AI in life sciences should treat data integrity and lab interoperability as core budget line items, not afterthoughts – because a faster model built on incomplete or manipulated data compounds risk rather than reducing it.

Originally reported by MIT Technology Review. Read the full story here.