What doctors need to know about AI's rising role in medical research

iStock.com/A_Pobedimskiy

By Michael Marshall

AI tools can review thousands of trials in days and spot promising drugs – but they also hallucinate, fabricate references and can quietly erode clinical judgment.

Barely a day goes by without another headline about a breakthrough in health research that was enabled, at least in part, by artificial intelligence (AI). In September, the journal Science published a study describing the Virtual Biotech: a cluster of up to 37,000 AI agents that can review published clinical trials and which identified a promising treatment for lung cancer. The AI agents may have saved thousands of person-hours of work.

Consensus is currently lacking around how to use AI in medical research – and when not to use it

Yet alongside this hype there are dire warnings. Generative AIs are known to hallucinate: that is, make stuff up. When researchers use AI to help them write their papers, the AIs can introduce errors. The rate of fabricated references in papers has risen tenfold over the last three years: in 2023 about one in 2828 papers had a fabricated reference but by early 2026 this was up to one in 277, according to a study led by Maxim Topaz at Columbia University in New York, published in The Lancet in May.