In January 2026, The BMJ published a study that should make anyone who hires, funds or promotes scientists uneasy. Researchers screened 2.65 million cancer-research papers published between 1999 and 2024. Their model flagged 261,245, or 9.87 per cent, as textually similar to papers already associated with paper mills.
That does not mean one in ten cancer papers is fraudulent. The authors did not claim that, and critics have warned that a textual classifier can produce false positives and language bias. What mattered was the trend. The signal rose sharply over time and appeared even in journals in the top decile by impact factor.
Counterfeit research can enter many of the same databases, journals and CVs as painstaking research. Once it does, the damage reaches beyond research integrity offices and into the labour market for scientists.
Science has built a Gresham problem.
The familiar shorthand for Gresham’s Law is that “bad money drives out good”. The missing condition matters: bad money drives out good when both are accepted at the same face value. If a debased coin and a full-weight coin buy the same loaf of bread, people spend the debased coin and keep the valuable one.
Academic hiring, promotion and funding create a rough version of that exchange rate. A paper becomes a line on a CV. A journal title becomes a quality signal. Publication counts and citations compress years of work into numbers a committee can process in an afternoon.
Some compression is unavoidable. Nobody reviewing 300 applications can reproduce every experiment. Trouble starts when the compression prices very different work too similarly.
Rigorous science is slow because reality keeps vetoing the story. Controls fail. Results weaken under replication. Analyses refuse to line up neatly. Each honest surprise costs time. A paper mill works under a different cost curve: fabricate or recycle data, manipulate an image, manufacture a manuscript, move to the next one.
If both outputs can earn comparable CV credit before anyone inspects the underlying evidence, the faster producer has a productivity advantage. Five weak or fraudulent papers can compete with one careful paper simply because the scoreboard can count to five more easily than it can assess four years of judgment.
That incentive does not require most scientists to become fraudsters. Most will not. Its effects begin much earlier.
A researcher can drop an awkward analysis, postpone the experiment most likely to break the hypothesis, tell a cleaner story than the data deserve, split one project into several papers or add a fashionable technique because reviewers have learned to expect it. None of these acts is equivalent to fabrication. They sit on the same incentive slope: decisions move away from “What would settle the question?” and toward “What will publish?”
The paper mill is the extreme case. The broader distortion begins wherever publishability starts choosing the science.
Researchers know what the market rewards. In a survey of faculty at 55 US and Canadian universities, respondents said they believed publication count, papers per year and journal name recognition were among the factors most valued in review, promotion and tenure. A separate survey of 6,813 researchers in the Netherlands found an association between publication pressure and frequent engagement in questionable research practices. More than half reported frequently engaging in at least one of the practices measured, although the authors cautioned that prevalence depends on definitions and survey design.
Those findings do not prove that metrics cause misconduct. They do establish the incentive environment in which misconduct, corner-cutting and publication optimisation compete with slower work.
Goodhart’s Law explains why a measure deteriorates once careers depend on hitting it. Gresham’s Law explains the next step: cheaper ways of producing the credential can start to outrun the expensive ones.
The publication toll
Luxury allocation systems offer a useful analogy. Buyers sometimes purchase ancillary goods partly to improve their chances of accessing the scarce item they actually want. The purchase has two purposes: the object itself, and admission to the next object.
Academic publishing can create the same behaviour around methods. A sequencing run may be essential. So may an omics dataset, an animal model or a difficult computational analysis. But a method becomes a publication toll when its main job is to make the manuscript look worthy of the journal rather than answer the scientific question.
A useful test is to ask whether the experiment would still be necessary if the target journal did not care whether the technique appeared in the paper. If not, some of the scientific budget is being spent on admission rather than discovery.
One researcher cannot fix this by opting out. If competing labs arrive with larger datasets, more techniques and longer publication lists, restraint can look like weakness. The arms race is individually rational even when it wastes time collectively. Early-career scientists understand this especially well: the laboratory can afford an aesthetic preference for simplicity; a CV often cannot.
Publishers sell what universities reward
Publishers did not invent promotion committees, grant scoring or the academic appetite for prestige. They have, however, become very good at supplying publication capacity to a market that prizes branded credentials.
Springer Nature reported that article output rose about 13 per cent in the first half of 2026, against estimated market growth of roughly 8 per cent. Its Research division generated EUR748 million of revenue in those six months, and the company launched 39 journals. In announcing the new Nature Progress series, Springer Nature said 84 per cent of surveyed authors welcomed more opportunities to publish in Nature-branded titles. The open-access article-processing charges were listed at EUR4,390 to EUR5,090.
Those numbers tell us nothing by themselves about whether editorial standards are falling. They do show that demand for branded publication capacity has commercial value. Universities make the wrapper valuable; publishers can then sell more wrappers.
The failure mode is visible in the Hindawi episode. Wiley-owned Hindawi journals retracted more than 8,000 papers in 2023 after investigations into systematic publication manipulation, and Wiley later retired the brand. Retraction notices described recurring problems that included discrepancies in research descriptions, unavailable data, inappropriate citations, incoherent content and compromised peer review.
Innocent authors pay too. Once a journal or publisher becomes associated with industrial manipulation, sound work carries a weaker signal simply because of where it appeared. Gresham’s Law damages holders of good currency too.
The impact-factor shortcut
The journal impact factor makes the exchange-rate problem worse because it invites committees to infer the value of an individual paper from the average performance of a journal.
Clarivate defines the metric at journal level: citations to recent content divided by the journal’s recent citable output. Citation distributions, however, are highly skewed. Nature Cell Biology noted that in journals it examined, roughly 65 to 75 per cent of articles received fewer citations than the journal’s impact factor.
Using that average to judge an individual paper is like assigning every company the average return of its stock index. The shortcut saves time. It also transfers prestige from a small number of heavily cited papers to every other paper carrying the same masthead.
That creates another arbitrage. If institutions reward the venue before they read the work, researchers have reason to optimise for acceptance into the venue. The journal name starts doing analytical work that properly belongs to the evidence.
China has already moved against parts of this metric culture. In 2020, its Ministry of Science and Technology instructed evaluators in certain basic-research programmes to use a representative-work system, generally limiting submitted papers to five and asking evaluators to focus on their quality and relevance. Education authorities also told universities to move away from rigid SCI publication requirements and cash rewards linked to papers. In March 2026, the National Science Library of the Chinese Academy of Sciences stopped publishing a journal-ranking list that had influenced research evaluation for 22 years.
Changing the metric does not end gaming. A new ranking can become the next target. Representative works can be stage-managed. Peer review can still be weak. But the direction is right because it makes papers less fungible. Five works that must be defended in substance are harder to mass-produce than fifty lines that merely have to be counted.
A serious evaluation system would go further. Reviewers could ask what the candidate personally contributed, which result would have falsified the central claim, whether the data are accessible, whether the work has been independently reproduced and how the authors handled corrections. Some first-pass assessments could hide journal names. Replications, negative results and durable datasets could count as outputs rather than acts of career charity.
Large institutions need shortcuts. The discipline is to keep the shortcut subordinate to the work it summarises.
Verification is now the scarce resource
The strongest defence of the current system deserves to be taken seriously. More retractions can mean detection is improving. More journals can widen access. Open access can move research beyond wealthy universities. Publishers are also investing in integrity systems. The STM Association says more than 35 publishers use its Integrity Hub to screen over 125,000 submissions a month, intercepting roughly 1,000 suspected paper-mill submissions monthly.
These systems are progress, but their scale shows how much of the cost has shifted into verification.
Digital publishing made distribution cheap. Paper mills, outsourced manuscript production and generative AI are making plausible-looking manuscripts cheaper to produce. Establishing that a paper deserves belief remains expensive. Someone has to inspect the images, trace the data, understand the methods, identify recycled text, test the statistics and, in the end, reproduce the result.
For much of modern academic publishing, acceptance carried information because both publication capacity and expert attention were scarce. Digital distribution weakened the first constraint without creating more expert attention.
As putting a paper online becomes trivial, proving what sits behind it becomes the expensive part. Data provenance, auditable workflows, serious post-publication review and credible replication become more valuable as manuscript production gets easier. Prestige survives only if it signals that verification happened rather than standing in for verification.
The same change has to reach hiring and funding. A paper should not receive career value merely because it exists in the right container. Committees need to price the underlying work: difficulty, contribution, provenance, reproducibility and what the result changed.
Every scientist eventually faces the private version of this problem. Late at night, after the clean figure has finally appeared on the screen, the uncomfortable question is simple: am I solving a problem, or am I minting currency?
One compromised choice can look trivial. Repeated across careers and institutions, those choices set the exchange rate for the field.
Gresham’s Law does not say bad money is intrinsically stronger than good money. It says bad money wins when the system insists on treating the two as equivalent. Science can change that exchange rate.
A publication should be valuable because of what survives scrutiny behind it, not because the wrapper once saved a committee time. Low-integrity paper will keep multiplying while the career market honours it at something close to face value. That is the exchange rate science has to change.