Researchers have developed a new artificial intelligence tool that identified more than 250,000 cancer studies for further review. The AI cancer studies screening system analyzed 2.6 million research papers published over the past 25 years. Scientists say the tool does not prove fraud but helps identify studies that deserve closer examination.
The findings, published in The BMJ, focus on research that shares writing patterns with papers linked to so-called “paper mills.” These businesses produce or sell scientific manuscripts and may use fabricated or manipulated data. However, researchers stressed that every flagged paper still requires careful review by human experts.
The research team at Queensland University of Technology (QUT) developed the tool using BERT, a machine-learning language model. They trained the system to recognize writing patterns found in previously identified paper mill publications. During testing, the tool correctly detected known paper mill papers about 91% of the time.
Researchers compared the system to an email spam filter. Instead of deciding whether a study is fraudulent, the tool simply identifies papers that may need additional checks. Three scientific journals have already started testing the technology as part of their editorial review process.
The study also revealed a sharp increase in potentially questionable cancer research. According to the researchers, the share of flagged studies rose from about 1% in the early 2000s to more than 16% by 2022. The highest numbers appeared in molecular cancer biology and laboratory-based research.
The findings come as publishers increase efforts to protect scientific integrity. Earlier this month, Nature reported that suspected paper mill studies often receive more citations than legitimate research. As a result, unreliable findings can spread through future scientific work and influence other researchers.
Questionable cancer studies may affect laboratory research, clinical trials, drug development, and treatment guidelines. They can also misdirect research funding and slow scientific progress. Therefore, publishers have introduced stronger peer review, plagiarism detection, and image analysis tools to improve research quality.
The researchers emphasized that artificial intelligence should support, not replace, human judgment. Editors will continue making the final decisions after reviewing flagged papers. As the use of generative AI continues to grow, the AI cancer studies tool could become an important safeguard for scientific publishing.
