A meticulous new study published in the Proceedings of the National Academy of Sciences on August 4 has warned that systematic scientific fraud is no longer a fringe concern but a pervasive, organised, and rapidly growing threat that jeopardises the foundations of research worldwide. The study has revealed a fine-grained break-up of the actors, methods, and scale behind industrialised academic misconduct.
Drawing on a trove of bibliometric and forensic data, the team — from the Northwestern University and the NSF-Simons National Institute for Theory and Mathematics in Biology, both in the US, and the University of Sydney in Australia — has revealed how coordinated entities like paper mills, brokerage firms, compliant editors, and unscrupulous journals work together to mass-produce fraudulent research.
In a personal blog post about the effort, Reese Richardson, the study’s lead author and a postdoctoral fellow at the Amaral Lab at Northwestern University, wrote, “The scientific enterprise is now witness to widespread, organised defection from the scientific public goods game. Large swaths of players, among them many scientists, reviewers, editors and publishers, are choosing to no longer make genuine contributions to the pot.”
A public goods game
The team framed its analysis using game theory, likening science to a sprawling public goods game in which progress is driven by collaboration, trust, and mutual investment. In the study’s framework, in exchange for generating knowledge and training the next generation, scientists receive societal rewards like funding and career advancement. However, as the size and complexity of science have both ballooned, so too have the incentives and opportunities to defect.
“While there has always been some concern that these pressures may compel some to defect from the scientific research ethos … the focus has largely been on the actions of lone individuals,” the team wrote in its paper. “Recently, however, reports of coordinated scientific fraud activities have increased”.
Richardson wrote that ‘defection’ was defined as “the act of choosing to contribute less than other players despite having the means to contribute”.
He added that in repeated public goods games simulated in the laboratory, players understand over time that defecting yields the greater advantage, leading to them contributing less and less to the collective pool. And although there is usually a group of players that cooperate to play the game in good faith, most players gradually lower their input. As a result, the total benefits from the pool dwindle while the number of defectors rises.
The paper also rationalised the use of the game theoretic framework as a means to analyse research misconduct as an organised activity rather than as errors committed by specific individuals: “Unethical behavior in science is often viewed as a character failure of an individual, not something perpetrated, enabled, and promoted by a cohort of individuals and entities. Indeed, even the definition of a now standard term such as ‘paper mill’ remains nebulous. Some of the organisations we describe may be better characterised as ‘brokerages’ than paper mills. We also cannot ascertain where our observations are due to the involvement of commercial paper mills or where they arise as a result of less formal peer networks operating on a noncommercial basis (as could be the case among some of the editors we flag).”
The authors added that the framework is also useful “because it frames some behaviour not in ethical terms but in terms of rationality. … For many junior doctors and budding scientists, engaging in defecting behavior may be the new norm.”
At the heart of this breakdown is the modern system of academic incentives. Funding and recognition increasingly hinge on quantitative proxies like publication and citation counts, h-indices, and journal impact factors, all of which can be artificially inflated.
Architecture of fraud
For their analysis, the team members used multiple data sources and analytic tools. Their sources included journal and article metadata from Clarivate’s Web of Science, Elsevier’s Scopus, PubMed/MEDLINE, and the OpenAlex databases — spanning several thousand journals and millions of articles — as well as lists of deindexed journals from the major indexing services and early-warning lists from Chinese oversight authorities. They combined this with data about retracted papers from the Retraction Watch database; metadata and content from PubPeer, a post-publication critique platform; and programmatic analyses of publisher data, notably from PLOS ONE and Hindawi, both of which label each article with its handling editor.
Upon analysis, the team found that certain editors at large journals, such as PLOS ONE and the stable of Hindawi journals, consistently handled disproportionately many articles that were eventually retracted or which received critical comments on PubPeer.
Using probabilistic modelling and statistical controls, the team could identify individuals whose pattern of acceptance couldn’t be explained by chance. These editors, many of whom also published each other’s work, formed tightly-knit clusters that, despite making up less than 1% of all editors, were implicated in most problematic articles at their journals.
One particular insight was that the fraud ecosystem has become resilient and adaptable. For example, as the paper put it, organisations such as the Academic Research and Development Association (ARDA) in India don’t only write and submit papers on behalf of clients but actively “journal hopped”, shifting its business to new journals as soon as existing venues were deindexed or scrutiny of its activities increased.
The team wrote that between 2018 and 2024, ARDA’s roster of guaranteed publication venues ballooned from 14 to more than 86 journals, including obscure or hijacked periodicals as well as journals indexed in Scopus, Web of Science, and MEDLINE. They added that the journals listed by ARDA have also been deindexed at rates vastly exceeding the baseline, often in apparent response to exposure events — although the deindexing also occurred too slowly to offset the tide of fraudulent output.
Evidence from journal archives has indicated that most articles published through ARDA’s network are beyond scope, with a significant share also representing improbable international collaborations. For example, the researchers found that of the five journals they comprehensively inspected from ARDA’s offerings, 10.1% of publications had authors from different countries; they also spotted a paper about roasting hazelnuts appearing in a journal about HIV/AIDS care. The team interpreted this to mean ARDA was selling papers’ authorships to the highest bidders.
Sobering numbers
An important plank of the analysis is the team’s construction of networks based on image duplication, which has become a hallmark of fabricated science. The researchers identified large clusters of articles published in the same journal, in the same year, and by the same publishers, all connected through shared or manipulated images. They were able to use statistical methods to show that this was a random occurrence: instead, the numbers are consistent with mass production and coordinated placement.
While all of science is susceptible, the extent of infiltration seems to be uneven. By comparing closely related subfields in RNA biology, Richardson et al. found that while error rates were similar across disparate new and expanding fields, the retraction rates differed dramatically. Subfields with formulaic, template-driven research, such as lncRNAs, miRNAs, and cancer, had retraction rates peaking at 4%, which significantly exceeded what the researchers said can be expected from honest error.
Perhaps the most sobering data exposed a mismatch between the scale of fraudulent output and the integrity of the mechanisms designed to address it. The corpus of suspected paper mill products has been doubling every 1.5 years, which the team has estimated is 10x faster than legitimate scientific publishing and far outpacing the growth of both retracted and flagged articles.
Even aggressive measures such as deindexing journals have been dwarfed by the sheer volume of compromised outlets. For example, fewer than 100 journals have been deindexed every year whereas there have been tens of thousands of journals and a staggering number of suspect publications.
According to the paper, “In response to concerns about editorial practices, [a few bibliometric aggregators] can deindex a journal. Web of Science and Scopus deindex on the order of a hundred journals each annually. While this may appear to be a large number, it is ten-fold smaller than the number of journals that publish paper mill products.”
“Extrapolating from current trends,” the paper added, “we estimate that only around 25% of suspected paper mill products will ever be retracted and that only around 10% of suspected paper mill products will ever reside in a deindexed journal.”
The winning strategy
The researchers also acknowledged some important limitations of their work. Foremost was that scientific fraud is by nature clandestine and even comprehensive data is not likely to accurately estimate its full scale. The patterns of detection and exposure are themselves biased by resources, attention, and field-specific vulnerabilities. Even so, the team wrote, the aggregate evidence “shows that the integrity of the extant scientific record and of future science is being undermined through the shortcomings in the very systems through which scientists infer the trustworthiness of each other’s work.”
The study and its accompanying reflections constitute both an urgent warning and a call for collective action within the scientific community. Industrialised scientific fraud is no longer a marginal concern, nor is it adequately deterred by current measures. Instead, the researchers have revealed a resilient ecosystem of actors who have been incentivised to defect repeatedly, by exploiting the metrics and weaknesses of the current system at the expense of honest research and scientific progress.
“These networks are essentially criminal organisations, acting together to fake the process of science,” the study’s senior author and Northwestern University professor of engineering sciences and applied mathematics Luís A. Nunes Amaral said in a statement. “Millions of dollars are involved in these processes.”
Without coordinated, better-resourced, and systematically independent approaches to detect, investigate, and sanction misconduct, the study’s findings show that the future of science is at risk of being shaped by those for whom defection is the rational way to go.