Science & Research Integrity

AI Flags More Than 250,000 Cancer Research Papers as Suspicious

When researchers turned an AI loose on 2.6 million cancer studies published over the last quarter-century, the results were alarming: more than 250,000 papers were flagged for signs of data manipulation, image duplication, and other questionable research practices. The findings, published in The BMJ, raise serious questions about the integrity of the scientific literature that doctors and policymakers rely on.

• The AI tool, developed by an international team of researchers, analysed 2.6 million cancer research papers published between 2000 and 2025. It flagged 253,669 papers — roughly one in ten — as containing indicators of problematic research. The tool does not label papers as fraudulent outright; it serves as a triage system, highlighting papers that warrant human review before they reach peer reviewers or clinical guidelines.
• Suspicious papers were heavily concentrated in specific fields. Molecular cancer biology and early-stage laboratory research had the highest flag rates, with some subfields seeing more than 20 percent of papers flagged. Certain cancer types — gastric, liver, bone, and lung cancer — were disproportionately affected. The pattern suggests that the pressure to publish high-impact results in competitive research areas may be driving the problem.
• The AI detected several types of misconduct indicators: duplicated or manipulated images (Western blots, microscopy images, and flow cytometry plots that appeared to have been reused or altered), implausible statistical results, and authorship patterns consistent with paper mill operations — factories that produce fake research papers for sale. Several major journals have already begun testing the AI tool as a pre-review screening mechanism.

Science is built on trust. When a researcher publishes a paper, the community assumes the data is real, the images are authentic, and the statistical analysis is honest. But over the past decade, a series of high-profile scandals — from the STAP stem-cell fraud to the growing paper mill industry — has eroded that trust. The scale of the problem has been difficult to measure because fraudulent papers are designed to evade human detection. This is where AI changes the game.

The tool developed by the team works by scanning papers for patterns that human reviewers rarely notice. It can detect when an image of a Western blot has been reused across different experiments, when a microscopy image has been rotated or flipped to appear new, and when statistical results are too perfect — a hallmark of data fabrication. In a pilot test, the AI flagged papers that had already been the subject of formal misconduct investigations, validating its accuracy. The system also identified authorship patterns that suggest paper mill involvement: authors with implausibly high publication rates, papers written in a style inconsistent with the authors' previous work, and co-author networks that span suspiciously broad geographic and institutional ranges.

The implications are sobering. If one in ten cancer research papers contains questionable data, then the scientific literature on which cancer treatments are based is partially contaminated. Clinical trials, meta-analyses, and systematic reviews that rely on the published literature may be drawing on flawed foundations. The AI tool offers a path forward: by screening papers before they enter the peer-review process, journals can catch problematic research before it becomes part of the permanent record. Several major publishers have already begun integrating the tool into their editorial workflows, and the team plans to extend the analysis to other medical fields. The fight against scientific fraud is entering a new era — one where AI is both the detective and the watchdog.