Modern medicine leans on a mountain of published research: every new treatment, every safety warning, every clinical guideline ultimately rests on studies that someone, somewhere, wrote up and shared. But how trustworthy is that mountain? A new AI-powered audit suggests the cracks are larger than many suspected.
The system swept through roughly 2.6 million cancer-research articles published between 1999 and the present, scanning titles and abstracts for linguistic and structural fingerprints associated with so-called "paper mills" — companies that mass-produce fake or ghostwritten manuscripts for authors who need publications but did not do the underlying science. Earlier work using large language models had already estimated that paper-mill-style cancer papers had climbed to around 15% of the literature in recent years. The new scan, covering far more records, points to an integrity problem at industrial scale rather than a handful of bad actors.
Why does this matter beyond academia? Cancer research feeds directly into drug approvals, treatment protocols, and the meta-analyses that clinicians rely on. If a meaningful slice of that foundation is fabricated, the whole tower of evidence can be quietly distorted — leading to wasted research, false leads, and, in the worst case, guidance built on results that were never real.
The useful twist is that the same technology causing concern is also part of the cure. Language models that can recognize the tell-tale patterns of manufactured papers offer journals and funders a scalable screening layer that human reviewers, facing thousands of submissions, simply cannot provide by hand. The goal is not to accuse individual scientists automatically but to triage the literature so suspicious work gets a closer look.
Knowledge takeaway: An AI audit of 2.6 million cancer papers since 1999 found integrity problems consistent with industrial-scale "paper mills." It is a warning about the reliability of the published evidence base — and a demonstration that AI, the suspected culprit, is also becoming the tool that helps clean it up.