AI Flags 250,000 Cancer Papers as Potential Fabrications

A new machine learning tool has identified more than 250,000 cancer research papers as potential fabrications ^[1][2]. The study, published in The BMJ, utilized a BERT-based language model to analyze 2.6 million studies released between 1999 and 2024 ^[2][3]. Researchers discovered that nearly 10% of these papers exhibited writing patterns consistent with fraudulent "paper mills" ^[1][3].

The prevalence of suspicious research has increased significantly over the last two decades. While fraudulent patterns were rare in the late 1990s, they reached a peak of 16% in 2022 ^[1][2]. This automated "scientific spam filter" highlights the growing challenge of maintaining data integrity in medical publishing ^[1][3].