AI hallucinations are inserting fabricated references into biomedical research at accelerating rates, infiltrating the scientific evidence chain that guides clinical treatment.

A study published in The Lancet audited nearly 2.5 million biomedical papers and 97 million citations on PubMed Central, finding more than 4,000 fabricated references across nearly 3,000 papers. The rate of fabricated references has grown more than 12-fold over the past three years: one in 2,828 papers in 2023, rising to one in 458 by last year, and one in 277 papers over the first seven weeks of 2026.

AI hallucinations occur when an AI model prioritizes word patterns over accuracy. Fabricated references embedded in biomedical papers can propagate through the scientific evidence chain—where clinical trials cite earlier studies, systematic reviews aggregate those trials, and medical guidelines cite those reviews—potentially compromising treatment decisions made by doctors and nurses.

Most papers contained only one or two fabricated citations out of several dozen total references, suggesting most AI hallucinations in research are unintentional. However, 98.4% of studies with fake references had not been retracted by publishers at the time of the audit, and verification methods differ widely between journals.