As scientific fraud proliferates, so do businesses that aim to stop it

Brokers of scientific fraud are constantly innovating and expanding. Companies looking to spot and tackle their activities are doing the same.

The last few years have seen a surge in activity from paper mills, which are shady entities looking to make a quick buck from publishing and spreading fraudulent science. Paper mills sell authorship slots on papers already slated for publication at scholarly journals or offer additional citations in exchange for a fee.

During the same period, multiple efforts have also been launched to identify, flag, and weed out these deceptive practices. Such projects—run by scholarly services firms, academic publishers, and the research community—look for a number of telltale signs that may indicate fraud. Because the area is a hot one, start-ups using technology to address the threats are gaining traction and attracting investment.

In March, for instance, ACS Publications, a division of the American Chemical Society, along with other angel investors, put $1.1 million into Signals, a London-based start-up founded with the goal of restoring the trustworthiness of academic literature. (C&EN is published by ACS.)

Elliott Lumb, cofounder of Signals, says that though paper mills were the center of attention in research integrity for a while, the focus has shifted to scientific fraud more broadly, especially given the rise of generative artificial intelligence. “It became very clear that the point is not to find paper mills,” he says. “It’s to find problematic and low-quality research that shouldn’t move forward to editors.”

Signals has developed software that looks for at least 30 trends that could indicate a paper is problematic. One of them is unlikely collaborations, where author lists include researchers based in several different countries and often working in entirely different disciplines. Others are articles that cited studies known to be affiliated with paper mills and papers containing text that is unrelated to the topic of the study.

“Because we have this underlying technology of the data graph layered with large language models linked into this, it’s actually becoming easier and easier for us to develop new and more robust signals of problematic research,” Lumb says. Signals, which has four full-time staffers and is hiring four more, started talking to ACS in mid-2025 not only about developing “deep evaluations” of research but also about how to determine whether papers actually add value and advance the status quo, he says.

How much publishers pay for Signals’ products varies according to the services they require, Lumb says. “We’re not just doing checks on a manuscript. We offer reports. We offer dashboards.”

Clear Skies, a start-up founded in 2022, claims to be the first commercial firm to release an AI tool designed to detect research fraud. Adam Day, the firm’s founder and CEO, demonstrated to C&EN a “risk rating” feature that monitors to what extent a journal’s papers are citing content that probably originated from a paper mill. “If someone cites dodgy stuff, it’s like the dodgy paper votes for them,” he says. “What we can show in our index is the journals that are not publishing bad stuff.”

Day worries about disclosing the names of journals with high-risk indices of fraud, noting that Clear Skies shares such data selectively. “Openness and responsibility are sometimes at odds,” he says. For example, although Clear Skies shows publishers data about their own titles, it won’t normally tell them about their competitors’ journals, Day says. The firm also works closely with the analytics firm Clarivate, which is trying to weed out bad actors from scholarly publishing.

Tools proliferate to tackle an entrenched problem

Other research integrity firms are surfacing, all offering their own specific services. For example, Grounded AI uses citation-context analyses to find paper mill activity, ReviewerZero AI is known for using large-language models to detect suspicious journals, and Pangram identifies text created by generative AI. Indexcat, which recently launched in beta, calculates a “content credibility index,” while ImageTwin tries to spot problematic images. A number of others exist.

To collate the surge of software and technology being developed in the area, in May 2022, the International Association of Scientific, Technical, and Medical Publishers (STM), a trade group representing academic publishers around the world, launched its research integrity hub. It contains tools, services, and widgets publishers can use to identify manuscripts that may originate from paper mills or generative AI and avoid wasting time and money refereeing fraudulent studies.

The effort has so far led to two applications that STM member publishers are using, says Joris van Rossum, program director at STM Solutions. The first allows publishers to manually upload their manuscripts to be screened by about 20 tools in the hub. The second automatically pushes papers from publishers’ editorial systems onto the hub and runs checks.

The hub screens over 200,000 manuscripts from more than 13,000 journals every month, van Rossum says. Publishers pay a fee that depends on the quantity of manuscripts. He notes that the STM doesn’t make suggestions about what publishers should do if papers in their journals are flagged by the hub.

Some of the tools in the hub were built by STM staff. Others were donated by publishers, provided by research integrity experts, or contributed by commercial firms, van Rossum says.

Some research integrity experts argue that many academic publishers aren’t doing enough to tackle the problems in their journals; in the case of open-access journals, they say, it’s because publishers collect article-processing charges (APCs) when they publish papers. But van Rossum thinks the reputational damage, plus the cost of screening papers and retracting them, outweighs the publishing fees. At least one publisher, the UK’s Institute of Physics Publishing, donates to charitable causes the APCs it earns from papers that are later retracted.

Fearing that he might tip off paper mill operators, van Rossum declines to reveal too many details about the tools. But they include a duplicate submission tool, which checks whether the same manuscript was submitted to multiple journals at the same time—a common modus operandi of paper mills. Other tools in the hub search for malfeasance by analyzing article references, text, digital object identifiers, and other metadata.

Also featured in the STM hub is the Problematic Paper Screener, which detects, among other indicators, tortured phrases—odd turns of phrase designed to disguise plagiarism of papers written in different languages—such as “abscisic corrosive” instead of “abscisic acid.”

Guillaume Cabanac, a computer scientist at the University of Toulouse who launched the Problematic Paper Screener in 2021, says paper mills are learning to avoid tortured phrases, perhaps with the aid of generative AI. But he’s still adding phrases that readers post on PubPeer, a site where scientists discuss papers.

Other indicators flagged by Cabanac’s screener include citations to retracted papers, papers published in hijacked journals (counterfeit websites mimicking real journals), studies mentioning problematic cell lines, and papers for which journals have issued notices of concern. Nonsensical studies created using the software SCIgen and Mathgen are also picked up.

As for the publishers dealing with all these problems, the response varies widely, according to Clear Skies’ Day. “Some publishers are really super engaged with this,” he says, “and some are not.”

Dalmeet Singh Chawla is a freelance science journalist based in London.