AI cancer research analysis flagged 261,000 papers (9.87%) as paper mill products, meaning nearly 1 in 10 published cancer studies may be fabricated

An AI-based analysis of 2,647,471 cancer research publications flagged 261,245 papers -- 9.87% -- as probable paper mill products. Separate estimates across biomedical research put the paper mill share between 2% and 20% of published papers, with one researcher warning that within a decade, more than half of annually published studies could be fraudulent. Wiley's own screening tool found that up to 1 in 7 submissions to hundreds of its journals showed signs of paper mill activity. The entities producing these papers are large, resilient, and growing rapidly, according to a 2025 PNAS study. In cancer research specifically, the consequences of paper mill contamination are not abstract. Oncologists designing clinical trials rely on published preclinical data to determine which drug targets, biomarkers, and treatment combinations to test in humans. If 10% of the preclinical literature is fabricated, researchers are building clinical trial designs on a foundation where roughly 1 in 10 supporting studies is fake. A Phase II clinical trial costs $10-50 million; a Phase III trial costs $50-300 million. Each trial designed around fabricated preclinical evidence wastes years of patient enrollment time, millions in funding, and -- most critically -- exposes cancer patients to experimental treatments justified by data that never existed. Failed trials also create 'negative evidence' that discourages future investigation of approaches that might actually work, because the field incorrectly concludes that the preclinical rationale was sound but the biology did not translate. Paper mills thrive because the incentive structure of academic publishing rewards publication volume, and the detection infrastructure cannot keep pace with production volume. Paper mills use AI to generate novel-looking western blot images, flow cytometry plots, and statistical tables that pass automated screening. They rotate author names, affiliations, and email domains to avoid pattern detection. They exploit the special issue model to bypass rigorous editorial oversight. And they serve a real market demand: researchers in systems where career advancement requires a minimum number of publications per year will pay $1,000-5,000 per paper to a mill rather than risk career stagnation. Until the incentive to publish-or-perish is decoupled from the incentive to fabricate, paper mills will continue to scale faster than detection tools.