Study Finds Nearly One in Ten Cancer Papers May Be AI-Generated Frauds
A machine-learning analysis identified that about 10% of cancer research articles published since 1999 show patterns typical of paper-mill or AI-generated fraud, and these papers are cited more often than genuine studies.
A team led by biostatistician Adrian Barnett used a BERT-based text classifier, trained on 2,202 confirmed paper-mill papers from the Retraction Watch database, to scan 2.6 million cancer research articles published between 1999 and 2024. The classifier, which reached 91% accuracy, identified more than 261,000 papers - about 10% of the sample - as bearing textual hallmarks of fraudulent, paper-mill or AI-generated origins.
The study highlighted a paradox: these suspect papers have garnered far more citations than authentic work, indicating that fraudulent findings are being incorporated into the scientific record. Similar contamination has been found in Alzheimer’s research, suggesting tens of billions of dollars have been invested in studies built on false premises. In response, France’s CNRS has ended costly subscriptions to major bibliographic services, revamped researcher evaluation, and deployed detection tools. Other analyses, such as a Northwestern University report in PNAS, show paper-mill output doubling every 1.5 years, far outpacing legitimate publication growth.
Why it matters
Fraudulent papers distort scientific knowledge and waste billions in research funding.
In this story
