Fake studies have flooded the publishers of top scientific journals, leading to thousands of retractions and millions of dollars in lost revenue. The biggest hit has come to Wiley, a 217-year-old publisher based in Hoboken, N.J., which Tuesday will announce that it is closing 19 journals, some of which were infected by large-scale research fraud.
In the past two years, Wiley has retracted more than 11,300 papers that appeared compromised, according to a spokesperson, and closed four journals. It isn’t alone: At least two other publishers have retracted hundreds of suspect papers each. Several others have pulled smaller clusters of bad papers.
Although this large-scale fraud represents a small percentage of submissions to journals, it threatens the legitimacy of the nearly $30 billion academic publishing industry and the credibility of science as a whole.
The discovery of nearly 900 fraudulent papers in 2022 at IOP Publishing, a physical sciences publisher, was a turning point for the nonprofit. “That really crystallized for us, everybody internally, everybody involved with the business,” said Kim Eggleton, head of peer review and research integrity at the publisher. “This is a real threat.”
Wiley will announce that it is closing 19 journals. Photo: Wiley
The sources of the fake science are “paper mills”—businesses or individuals that, for a price, will list a scientist as an author of a wholly or partially fabricated paper. The mill then submits the work, generally avoiding the most prestigious journals in favor of publications such as one-off special editions that might not undergo as thorough a review and where they have a better chance of getting bogus work published.
World-over, scientists are under pressure to publish in peer-reviewed journals—sometimes to win grants, other times as conditions for promotions. Researchers say this motivates people to cheat the system. Many journals charge a fee to authors to publish in them.
Problematic papers typically appear in batches of up to hundreds or even thousands within a publisher or journal. A signature move is to submit the same paper to multiple journals at once to maximize the chance of getting in, according to an industry trade group now monitoring the problem. Publishers say some fraudsters have even posed as academics to secure spots as guest editors for special issues and organizers of conferences, and then control the papers that are published there.
“The paper mill will find the weakest link and then exploit it mercilessly until someone notices,” said Nick Wise, an engineer who has documented paper-mill advertisements on social media and posts examples regularly on X under the handle @author_for_sale.
The journal Science flagged the practice of buying authorship in 2013. The website Retraction Watch and independent researchers have since tracked paper mills through their advertisements and websites. Researchers say they have found them in multiple countries including Russia, Iran, Latvia, China and India. The mills solicit clients on social channels such as Telegram or Facebook, where they advertise the titles of studies they intend to submit, their fee and sometimes the journal they aim to infiltrate. Wise said he has seen costs ranging from as little as $50 to as much as $8,500.
When publishers become alert to the work, mills change their tactics.
[…]
For Wiley, which publishes more than 2,000 journals, the problem came to light two years ago, shortly after it paid nearly $300 million for Hindawi, a company founded in Egypt in 1997 that included about 250 journals. In 2022, a little more than a year after the purchase, scientists online noticed peculiarities in dozens of studies from journals in the Hindawi family.
Scientific papers typically include citations that acknowledge work that informed the research, but the suspect papers included lists of irrelevant references. Multiple papers included technical-sounding passages inserted midway through, what Bishop called an “AI gobbledygook sandwich.” Nearly identical contact emails in one cluster of studies were all registered to a university in China where few if any of the authors were based. It appeared that all came from the same source.
[…]
The extent of the paper mill problem has been exposed by members of the scientific community who on their own have collected patterns in faked papers to recognize this fraud at scale and developed tools to help surface the work.
One of those tools, the “Problematic Paper Screener,” run by Guillaume Cabanac, a computer-science researcher who studies scholarly publishing at the Université Toulouse III-Paul Sabatier in France, scans the breadth of the published literature, some 130 million papers, looking for a range of red flags including “tortured phrases.”
Cabanac and his colleagues realized that researchers who wanted to avoid plagiarism detectors had swapped out key scientific terms for synonyms from automatic text generators, leading to comically misfit phrases. “Breast cancer” became “bosom peril”; “fluid dynamics” became “gooey stream”; “artificial intelligence” became “counterfeit consciousness.” The tool is publicly available.
Another data scientist, Adam Day, built “The Papermill Alarm,” a tool that uses large language models to spot signs of trouble in an article’s metadata, such as multiple suspect papers citing each other or using similar templates and simply altering minor experimental details. Publishers can pay to use the tool.
[…]
The incursion of paper mills has also forced competing publishers to collaborate. A tool launched through STM, the trade group of publishers, now checks whether new submissions were submitted to multiple journals at once, according to Joris van Rossum, product director who leads the “STM Integrity Hub,” launched in part to beat back paper mills. Last fall, STM added Day’s “The Papermill Alarm” to its suite of tools.
While publishers are fighting back with technology, paper mills are using the same kind of tools to stay ahead.
“Generative AI has just handed them a winning lottery ticket,” Eggleton of IOP Publishing said. “They can do it really cheap, at scale, and the detection methods are not where we need them to be. I can only see that challenge increasing.”
