Researchers at Queensland University of Technology in Australia built a machine-learning system that compares the writing style, abstract structure and phrasing of medical papers against a known database of fraudulent studies. They ran it across 2.6 million cancer research papers published between 1999 and 2024. The result: more than 250,000 papers — roughly one in ten of everything published on the topic over a quarter-century — carried writing patterns that match those produced by so-called "paper mills."
A paper mill is an operation that churns out low-quality or outright fabricated manuscripts for sale, often stuffing them with recycled sentences, invented statistics and fake image figures. They exploit the sheer volume of submissions that journals must process; once a paper clears editorial and peer review, its findings enter textbooks, drug-trial planning and clinical guidelines. If the foundation is contaminated, every downstream decision built on it is suspect.
The scale of the finding is what alarms researchers. Cancer science is one of the most citation-heavy and commercially consequential fields in medicine — a single influential study can redirect millions of dollars in research funding or shape how a pharmaceutical company prioritises a therapy. A screening tool that flags one in ten papers suggests the problem is no longer a handful of rogue journals but a systemic pressure: publish-or-perish incentives, overloaded reviewers, and a flood of submissions have made the review process harder to defend.
Crucially, the AI does not accuse any given paper of fraud. It tags papers as "suspicious" so that editors, institutions and retraction-watch groups can prioritise human investigation where the risk is highest. Think of it as a metal detector at an airport: it raises an alarm, but a human still decides whether the trip is real. The tool is already being discussed as a routine pre-publication screen that journals could run on every incoming submission, turning peer review from a random lottery into a risk-stratified process.
Knowledge takeaway: A QUT machine-learning screen compared 2.6 million cancer papers (1999–2024) against known fraudulent writing patterns and flagged more than 250,000 — about one in ten — as suspicious; the flagged papers share traits of "paper mills," operations that sell fabricated or low-quality manuscripts; the tool is intended as a triage aid that lets journals focus human review on high-risk submissions rather than a verdict of guilt in any single case.