Scientific misconduct is not a fringe problem. A comprehensive analysis of more than 16,000 retracted medical publications spanning 1975 to 2024 found that fraud and data concerns are the leading drivers of retractions, and both categories have been roughly doubling every five years. The scale and variety of unethical research have shifted in the past decade, from lone fabricators to organized paper mills, from crude data falsification to AI-generated datasets, and from isolated informed-consent violations to entire commercial industries selling unproven treatments. What follows is a look at the specific forms this misconduct takes, who it harms, and why the systems meant to prevent it keep falling short.
The Retraction Epidemic
Retractions used to be rare events, typically tied to honest errors caught after publication. That picture has changed dramatically. A landmark review of all 2,047 biomedical and life-science articles listed as retracted in PubMed by mid-2012 found that only about a fifth were attributable to error. Roughly two-thirds were retracted because of misconduct: fraud or suspected fraud accounted for about 43%, duplicate publication for 14%, and plagiarism for close to 10%. The same study noted that the share of papers retracted for fraud had increased roughly tenfold since 1975.1PubMed Central. Misconduct accounts for the majority of retracted scientific publications
More recent data confirm the trend has only accelerated. An analysis of 16,041 retracted medical documents from 1975 to 2024, drawing on the Retraction Watch database, found that data concerns topped the list at about 31%, followed by fraud at around 11%, peer-review manipulation at 11%, referencing issues at roughly 8%, and ethical violations at 7%. The doubling times are striking: retractions linked to data concerns doubled roughly every five and a half years, and fraud-related retractions doubled about every five years.2PubMed Central. Fifty Years of Retracted Medical Publications From 1975 to 2024: A Comprehensive Analysis of Trends, Reasons, and Countries Using the Retraction Watch Database Part of that rise reflects better detection rather than a pure increase in wrongdoing, but the sheer volume suggests that detection alone does not explain everything.
Paper Mills and Industrial-Scale Fraud
One of the most alarming developments of the past decade is the emergence of paper mills. These are commercial operations that produce fake manuscripts for profit. They fabricate data, generate figures, and sell authorship slots to researchers who need publications for career advancement. The result is that many thousands of fraudulent papers have made it into peer-reviewed journals.3PubMed. Paper mill challenges: past, present, and future Unlike traditional misconduct, where a single researcher falsifies results, paper mills operate at an industrial scale, sometimes producing hundreds of manuscripts at a time on topics their authors may know little about.4Meta-Psychology. Paper mills: a novel form of publishing malpractice affecting psychology
The problem has drawn regulatory attention in multiple countries. China, which produces a large share of global research output, introduced misconduct rules specifically targeting paper mills.5PubMed. China’s research-misconduct rules target ‘paper mills’ that churn out fake studies But the incentive structure that fuels paper mills is global. Researchers in many countries face career systems that reward publication volume above nearly everything else. When promotions, tenure, and grant renewals hinge on publishing a certain number of papers per year, some fraction of researchers will look for shortcuts, and paper mills exist to sell them one.
Paper mills also create a secondary problem: they pollute the evidence base. When fabricated studies on drug interactions, toxicology, or surgical outcomes sit in the literature undetected, they can inform clinical guidelines, meta-analyses, and public-health decisions. A single fake paper in an obscure journal may seem harmless, but when thousands accumulate, they degrade the reliability of the entire system.
AI as an Accelerant
Artificial intelligence has made paper-mill operations cheaper and harder to catch. Large language models can produce plausible-sounding manuscripts at speed. AI tools can fabricate synthetic datasets that pass basic statistical checks, generate fake references that look real until someone tries to look them up, and rewrite plagiarized text so thoroughly that standard plagiarism detectors miss it.6International Journal of Public Research in Medicine and Health. Artificial Intelligence–Driven Scientific Misconduct: Synthetic Data, AI-Generated Manuscripts, And Emerging Threats to Research Integrity—A Narrative Review The concern is not hypothetical. Journals have already flagged papers containing hallucinated citations, nonsensical AI-generated text left unedited, and datasets with statistical properties that suggest they were never collected from actual experiments.
The integrity challenges posed by AI extend beyond paper mills. Citation cartels, in which groups of authors or journals artificially inflate one another’s citation counts, have become easier to orchestrate when bots can generate reviews and cross-references automatically. Fake reviewer identities can be created at scale to manipulate peer review.7PubMed. Addressing New Challenges to Scientific Integrity Including Paper Mills, Citation Cartels, Coercive Citation Practices, Fake Reviewers, and Artificial Intelligence-Generated Papers The tools that were supposed to democratize scientific communication have also democratized fraud.
Image Manipulation in Published Research
Fabricated or duplicated images have become one of the more visible forms of research misconduct, in part because they leave behind forensic traces that are easier to detect than statistical fabrication. A screening of more than 20,000 papers published across 40 journals from 1995 to 2014 found that about 3.8% contained problematic figures, and at least half of those showed signs of deliberate manipulation rather than accidental error. The rate of problematic images rose over the period studied.8PubMed Central. The Prevalence of Inappropriate Image Duplication in Biomedical Research Publications
In some journals and fields, the numbers are far worse. A focused analysis of the journal Toxicology Reports found that 16% of papers containing relevant images had inappropriate duplications. Notably, careful manual reading of the papers and their cited literature could have caught most of these before publication, suggesting a failure of peer review rather than just author misconduct. Adding AI-based screening software increased the detection rate substantially, catching 41 problematic papers that manual screening alone had missed.9bioRxiv. A Quantitative Study of Inappropriate Image Duplication in the Journal Toxicology Reports Automated detection frameworks using deep learning methods have shown accuracy rates around 90% for identifying duplicated images, which is considerably better than what human reviewers achieve alone.10arXiv. Learning to identify image manipulations in scientific publications
Clinical Trials and Vulnerable Populations
Ethical breaches in human-subjects research are not confined to the historical abuses most people know about. Contemporary concerns center on the globalization of clinical trials, particularly in low- and middle-income countries where participants may have little access to healthcare outside the trial itself. Critics have pointed out that participants in these settings are often poorly educated about the nature of the research, may not fully understand what informed consent means, and are vulnerable to coercion because the trial offers medical care they could not otherwise afford.11PubMed Central. The ethics of global clinical trials: In developing countries, participation in clinical trials is sometimes the only way to access medical treatment. What should be done to avoid exploitation of disadvantaged populations?
The ethical problem is not that clinical trials occur in developing countries. Excluding those populations from research would itself be unjust, since it would mean drugs are tested only on wealthy populations and then prescribed globally. The problem arises when the standards of consent, compensation, and post-trial access that would be required in a wealthy country are quietly lowered for convenience or cost savings. When a trial ends and the experimental drug turns out to work, do participants get continued access? Too often the answer is no, which raises the question of whether the trial relationship was genuinely reciprocal or simply extractive.
Gene Editing Without Guardrails
In 2018, a Chinese researcher announced that he had used CRISPR-Cas9 to edit the genomes of twin embryos before implantation, making them the first known gene-edited humans. The announcement drew swift international condemnation, not because gene editing itself is inherently unethical, but because the experiment leapfrogged the global consensus that germline editing in humans should not proceed until the safety and ethical questions are resolved.12Bangladesh Journal of Bioethics. CRISPR/Cas9-mediated Editing of Human β-globin Gene in Human Cells: A Commentary on the Research Ethics
Germline editing changes DNA that gets passed to future generations, which raises a cascade of problems. Undesirable off-target mutations could appear not just in the edited individual but in their descendants. Meaningful informed consent is impossible to obtain from people who do not yet exist. And the technology creates a clear path toward selection for traits unrelated to disease, raising the specter of modern eugenics.13PubMed Central. Bioethical issues in genome editing by CRISPR-Cas9 technology The researcher responsible was eventually sentenced to prison, but the episode revealed how little enforcement power the international scientific community has when a researcher in one country decides to push past an ethical boundary that others consider inviolable.
The Unregulated Stem Cell Industry
While gene editing grabbed headlines for its dramatic boundary-crossing, a quieter and arguably more widespread ethical problem has been growing in plain sight. By March 2021, researchers identified 1,480 businesses in the United States alone operating 2,754 clinics selling purported stem cell treatments. That represented more than a fourfold increase in just five years. These businesses sell products that are not approved by the FDA and lack convincing evidence of safety or efficacy.14Cell Stem Cell. The American stem cell sell in 2021: U.S. businesses selling unlicensed and unproven stem cell interventions
These clinics market directly to consumers, often using testimonials and scientific-sounding language to promote treatments for conditions ranging from arthritis to neurological diseases. Patients pay thousands of dollars out of pocket for procedures whose true content and effectiveness have not been established.15PubMed Central. Rogue stem cell clinics The problem is global, and coordinated regulatory action has been proposed but has proven difficult to implement.16PubMed. Marketing of unproven stem cell-based interventions: A call to action Some patients have suffered serious harm, including blindness after injections of unproven preparations into the eye. The stem cell clinic industry sits in an uncomfortable gray zone: it is not traditional research misconduct, since the operators rarely publish fraudulent papers, but it exploits the language and authority of science to sell unvalidated treatments to desperate people.
Data Scraping and Consent in Digital Research
Research ethics used to be almost entirely about physical experiments on human or animal subjects. The digital age has created new categories of ethical violation that existing regulations were never designed to address. Facial recognition technology offers a vivid example. Clearview AI built its reference database by scraping user profiles from Facebook, an action explicitly prohibited by the platform’s terms of service. IBM collected individuals’ Flickr photos without their consent and did not give affected users a feasible way to have their images removed.17PubMed Central. The ethics of facial recognition technologies, surveillance, and accountability in an age of artificial intelligence: a comparative analysis of US, EU, and UK regulatory frameworks
These cases blur the line between corporate product development and research. The datasets built from scraped images are used to train algorithms that end up in law enforcement, border control, and commercial surveillance. People whose faces appear in training data never consented to that use and typically have no recourse. Traditional research ethics frameworks, built around the model of a scientist asking a participant to sign a consent form, do not map cleanly onto a world where billions of data points are harvested passively.
Gain-of-Function Research and Biosecurity
Some of the most contentious ethical debates in recent years have surrounded gain-of-function research, experiments that deliberately enhance the transmissibility or virulence of pathogens in order to study pandemic preparedness. Proponents argue that understanding how a virus could become more dangerous is essential for developing vaccines and surveillance tools. Critics counter that the experiments themselves could cause a pandemic if a modified pathogen escaped the lab. In 2014, the Obama administration called for a pause on federal funding for gain-of-function experiments involving influenza, SARS, and MERS viruses.18PubMed Central. Gain-of-Function Research: Ethical Analysis
The debate has resurfaced with added urgency. The current U.S. administration has proposed a renewed pause on domestic gain-of-function research while policies are revised. But as critics of the pause have pointed out, domestic restrictions cannot compel other nations to stop. Halting the work unilaterally may simply mean that countries with less transparent oversight programs continue the research without the safety standards that U.S. labs would apply, and the resulting knowledge gap could itself be a security risk.19mSphere. Dual use and gain-of-function research: a significant endeavor with biosecurity imperatives The ethical tension here has no clean resolution: the research is dangerous, but ignorance about pathogen evolution may be dangerous too.
Publish or Perish and the Economics of Misconduct
Behind many of these problems sits a structural incentive. Academic careers in most countries depend heavily on publication output. Hiring committees count papers. Promotion committees count citations. Grant agencies look at track records measured largely in published work. This “publish or perish” system has long been identified as a driver of misconduct: increased research productivity worldwide has been accompanied by a rise in falsified and erroneous data that can mislead clinicians and policymakers into allocating resources based on fabricated evidence.20PubMed Central. Research Misconduct: The Peril of Publish or Perish
Conflicts of interest add another layer. Pharmaceutical and medical device companies routinely pay researchers, and transparency regulations in many countries now require those payments to be disclosed. But compliance is uneven. A scoping review found that the rate of agreement between what authors self-reported about industry payments and what companies actually reported ranged wildly, from as low as 1% in one study of ophthalmology journals to 94% in a study of surgical trauma researchers. Journal articles showed the lowest median compliance overall.21PubMed Central. Research on policy mechanisms to address funding bias and conflicts of interest in biomedical research: a scoping review When industry funding goes undisclosed, readers cannot judge whether a finding might have been shaped by the funder’s interests. The result is not always outright fraud, but it erodes the trust that makes science useful.
Predatory Journals and the Misinformation Pipeline
Predatory journals occupy a niche that makes all of the above worse. These publications claim to be legitimate scholarly outlets but misrepresent their practices. They falsely claim to provide peer review, hide information about article processing charges, and sometimes list editorial board members without their knowledge or consent.22PubMed Central. Predatory Journals: What They Are and How to Avoid Them For researchers who need publications quickly, predatory journals offer a fast path to a line on a CV. For paper mills, they provide a low-friction outlet. And for the public, they create a pipeline through which bad science enters the information ecosystem and gets amplified.
Predatory journals are one of at least four channels through which health misinformation originates within the scientific system itself. The others include pseudoscientists who use scientific-sounding language without adhering to the scientific method, legitimate scientists who spread inaccurate information (sometimes inadvertently, sometimes not), and media outlets that miscommunicate scientific findings.23PubMed Central. Reducing Health Misinformation in Science: A Call to Arms What makes predatory journals particularly insidious is that they wear the trappings of legitimate science. A paper published in a predatory journal has a DOI, appears in some databases, and looks indistinguishable from real research to anyone who does not know to check the journal’s credentials.
Dishonesty in Dishonesty Research
Perhaps the most ironic trend in recent years has been the discovery of data fabrication in studies about human honesty. A body of prominent behavioral science research on lying, cheating, and moral behavior has come under scrutiny after post-publication analysis revealed suspicious patterns. One systematic check of the dishonesty research literature found that 11 out of 61 articles containing multiple statistical tests produced results that were “too good to be true,” meaning the reported findings showed less variability than would be expected from real data.24Meta-Psychology. Untrustworthy Evidence in Dishonesty Research Some of the most widely cited studies on nudging people toward honesty turned out to rest on fabricated data, an outcome so perfectly ironic that it almost reads as satire. But the consequences are real: institutions and companies had already adopted interventions based on those findings.
Post-Publication Watchdogs and Their Limits
The most effective checks on misconduct increasingly come not from journals or institutions but from post-publication platforms where anonymous scientists flag problems. PubPeer, the most prominent of these, allows users to comment on published papers, pointing out image duplications, statistical anomalies, or plagiarism. Post-publication peer review has been described as an essential complement to the initial review process, enabling a broader set of experts to identify flaws that the original reviewers missed.25PubMed Central. An expert criticism on post-publication peer review platforms: the case of pubpeer
But PubPeer’s success has created its own problems. The volume of comments flagging potential misconduct has overwhelmed the institutions responsible for investigating them. Research misconduct regulations in the United States were written two decades ago, before anyone anticipated that an anonymous online platform would generate a steady stream of allegations. Universities, legally obligated to investigate each one, have found their research integrity offices swamped, diverting attention and resources from other integrity initiatives.26PubMed. The PubPeer conundrum: Administrative challenges in research misconduct proceedings The bottleneck is not detection anymore. It is response.
Why Misconduct Rarely Faces Criminal Consequences
One reason the system struggles to deter misconduct is that it almost never leads to criminal prosecution. The scientific community has been largely successful in keeping criminal law away from misconduct cases. Investigations are typically handled by committees of scientists, often from the same institution as the accused, in proceedings that frequently lack transparency.27Research Ethics. Should research misconduct be criminalized? The worst outcomes for a researcher found guilty of misconduct tend to be retraction of papers, loss of funding, and occasionally job loss. For someone who built an entire career on fabricated data, those penalties may arrive years after the damage was done and after the fraudulent findings have already influenced clinical practice or public policy.
Advocates for criminalization argue that when fabricated research leads to patients receiving ineffective or harmful treatments, the responsible researcher has caused a public harm comparable to fraud in other regulated industries, and the consequences should reflect that. Opponents worry that criminal penalties would chill legitimate risk-taking in science and push misconduct investigations further underground. The debate remains unresolved, but the status quo, in which the penalties for misconduct are mild and the investigations opaque, does not appear to be slowing the trend.