Cancer · Prediction and risk
AI flags nearly 1 in 10 cancer papers as looking like paper-mill fakes
A BMJ study trained an AI on 2,202 retracted fake papers and used it to screen 2.6 million cancer studies. It flagged 261,245 of them, almost 1 in 10, as resembling paper-mill work. Flagged means 'looks similar', not 'proven fake'. Rating: Early research.
How far away is this?
A peer-reviewed screening study. It's a signal, not proof about any single paper.
Get the weekly How Far
One email every Friday morning (Sydney time) on which medical AI actually reached patients, and what it means for you. Free, in plain English.
Where is it approved?
| Region | Regulator | Status | Notes |
|---|---|---|---|
| United States | FDA | Not applicable | Research tool; not a medical device. |
| European Union | CE mark / EMA | Not applicable | No register record checked. |
| United Kingdom | MHRA | Not applicable | No register record checked. |
| Australia | TGA | Not applicable | No ARTG record checked. |
Not a medical product. "Register not searched yet" means we haven't searched that regulator's register; it doesn't mean the product is or isn't available. Approval in one country isn't approval in another.
What is it?
Paper mills are businesses that sell made-up or recycled research papers. The AI read titles and abstracts and scored how much each paper looked like known fakes.
What's the evidence?
The model flagged 261,245 of 2,647,471 cancer papers from 1999 to 2024 (9.87%). That's about 1 paper in every 10.[1]
Evidence stage: Early research · #c1
It got 91% of test papers right: it caught 87% of known fakes and correctly cleared 96% of genuine papers. Those figures come from a test set of known retracted papers; on a separate set of papers from research-integrity experts it scored 93% (catching 87% of fakes and clearing 99% of genuine papers). Accuracy on other kinds of paper-mill work is unknown, and the AI only read titles and abstracts, not full papers or data.[1]
Evidence stage: Early research · #c2
The authors agree the 9.87% shouldn't be read as the true rate of fraud; it's an approximate signal.[2]
Evidence stage: Early research · #c3
What does it mean for me?
Treatment guidelines lean mainly on large clinical trials and reviews, which are harder to fake than small lab papers, but fake studies can still pollute the evidence base. That's why replication and big, independent trials matter. This doesn't mean any particular paper, or your treatment, is fraudulent.
What could I ask my doctor?
- Is my treatment backed by large clinical trials?
These are general starting points, not a recommendation to choose any particular test or treatment.
Short videos, under a minute
We make short, plain-English videos about AI in medicine, including what it can't do yet. Follow How Far for the next one.
Change history
- : First published.
- : Editor pass before publication: labels aligned to site categories; sources and rating reviewed.
General information only, not medical advice. Talk to your doctor about your situation. AI-assisted source check. Not clinically reviewed.
References
- Scancar et al., BMJ (30 Jan 2026). doi.org (2026). Peer-reviewed study. https://doi.org/10.1136/bmj-2025-087581
- Authors' reply to rapid responses. bmj.com. Expert commentary. https://www.bmj.com/content/392/bmj-2025-087581/rr-7
Publisher shown as the website hosting each source. Dates are given where the source label includes them.