Fabrication, fraud and their detection
Fabrication and falsification create an untrue research record, and detecting them may require evidence beyond an individual manuscript or its peer review.
Forms and prevalence
Scientific misconduct includes fabrication, falsification and plagiarism. Fabrication reports invented results as genuine findings. Falsification manipulates research materials, equipment, processes, data, information or citations and can include omitting observations to improve apparent means or trends. Excluding outliers can turn a nonsignificant result into a statistically significant one. A 2012 account used fraudulent publication as an umbrella for plagiarism, fabrication and falsification and distinguished misconduct from honest error or honest differences of interpretation.
A 2009 synthesis of 21 self-report surveys found that fewer than 2% admitted fabrication or falsification and nearly 34% admitted other questionable research practices; reports about colleagues were higher, reaching 72% for other questionable practices. The synthesis regarded these rates as likely underestimates because the questions were sensitive. Discrepant replication results do not by themselves establish fabrication, because methods and experimental limitations may differ, and most falsified data tend to be discovered by chance.
Limits of peer review
An experienced editor argued in 2007 that peer review was not designed to detect fraud: reviewers were chosen for scientific expertise and lacked the time, resources and investigative remit needed to uncover it. Passing review therefore provides no meaningful evidence that fabrication has been excluded. Published accounts also suffer from ascertainment bias: journals can reject suspicious work privately on less evidence than would justify a public accusation.
Two Science papers in 2004 and 2005 claimed embryonic stem-cell lines from cloned human embryos; the 2005 paper underwent internal and external review before rapid acceptance and online publication. After questions about egg donation, data handling and interpretation, a Seoul National University panel concluded that necessary supporting data had been fabricated, and the papers were retracted in early 2006. The journal’s editor held that skilfully crafted fraud could evade peer review.
At the Journal of Laboratory and Clinical Medicine, a reviewer and reader questioned an author’s unusually numerous single-author patient-follow-up papers, suspecting fabrication or denial of colleagues’ credit. Repeated institutional moves left no academic authority to investigate, so analysis of the author’s 49-paper corpus revealed impossible sequences, including human work preceding its supposed animal basis and studies preceding the proposed mechanism. The journal published its concerns beside the author’s response, judged that response insufficient and withdrew support for its papers. The individual papers had looked credible; inconsistencies emerged when they were considered together, and withdrawal did not undo the damage caused by publication or stop later citation.
John Darsee’s credible-looking papers likewise escaped reviewers and senior collaborators until someone noticed reports involving more animals than he had purchased. The editor suggested that trust in a promising trainee could impede detection and that reviewers’ constructive advice could help a fabricator produce more convincing invented studies. He proposed effective but proportionate institutional oversight, author certification of relevant skills and facilities, and a time-limited commitment to retrospective reanalysis, perhaps five years, supported by data banking. Such access could support alternative analyses as well as investigation of alleged irregularities.
Image manipulation and scrutiny
In 2012, an American Physiological Society publications representative advised against moving, removing, introducing, obscuring or enhancing any image feature; a blot should appear as captured, including blemishes. She described image manipulation as the most prevalent misconduct in APS submissions, whether authors knew alteration was disallowed or not. APS received about 8,000 manuscripts annually and projected 161 misconduct investigations in 2012, with cases increasing. She identified three publishing-environment risk factors.
- Author inexperience or lack of mentoring in appropriate standards.
- Availability of figure-manipulation software.
- Increased competition for positions and funding.
A study of 463 MEDLINE retractions from 1966–2008 attributed 213 to author misconduct, most commonly plagiarism and data falsification or fabrication. In 2012, multiple papers from one Japanese laboratory in Circulation, Kidney International and Hypertension were reported to reuse immunofluorescence and histology images and Western and Northern blot bands; some journals had issued expressions of concern, with retractions still anticipated.
Systematic reviewers can encounter suspect data: among 118 reviews analysed in 2013, seven suspected misconduct but only two reported it explicitly. The suspicions comprised falsification in three reviews, manipulation in one, a discrepancy between a publication and reanalysed FDA-posted data in one, and selective outcome reporting in two.
AI and the scale of fabrication
A publisher representative reported in 2024 that generated text and images made fabrication faster and easier and that the publisher had expanded its ethics team substantially, principally to address paper mills. Another representative regarded fraudulent authors and papers as a small minority and suggested AI might increase existing fraudsters’ output rather than greatly increase their number, with patterns of author behaviour helping detection. The panel’s concluding view was that generative AI amplified existing integrity problems whose effects on publishing were disproportionate to their prevalence.
Paper mills predated generative AI; one 2024 panellist linked them to publishers’ incentives for larger journals and increased output. Distinguishing fraudulent scholarly content in AI outputs can be difficult when the output does not reveal which sources it draws on.
Examples discussed in 2024 included a published AI-generated rat-anatomy image with fabricated cell-signalling diagrams and a paper retaining an introduction-generation instruction, illustrating failures to catch such content before publication. A JAMA representative described plans for image-integrity checks, detection of AI-generated or paper-mill submissions and AI-assisted triage, with human oversight remaining necessary.
Reporting allegations and consequences
EMA’s policy for external allegations of impropriety took effect on 17 March 2017, covering departures from good practice that could affect medicine evaluation or supervision and protecting reporters’ confidentiality. Prepared with the European Commission and OLAF, it provided email and postal reporting routes and structured assessment and investigation. EMA coordinated investigations concerning centrally authorised medicines, with possible scientific-committee action over benefit–risk concerns, and could refer nationally authorised cases to the relevant member-state agency; suspected fraud went to OLAF. By the policy’s introduction, EMA had received 43 reports since 2013, including manufacturing and clinical-trial concerns, already handled under its principles.
In 2014, Dong-Pyou Han was prosecuted over alleged fabrication of HIV-vaccine experiments at Iowa State University, after resigning in autumn 2013 and accepting three years’ exclusion from federal funding; he initially denied the charges. The experiments had appeared to show a strong antibody response to an HIV glycoprotein. NIH had awarded $5 million on applications and reports partly relying on the disputed data, and the university agreed to repay nearly $500,000 associated with his salary. After a guilty plea, Han received a 57-month prison sentence on 1 July 2015 and an order to repay $7.2 million to NIH, with deportation considered probable at that time. A Nature.com poll found divided opinion over the punishment, and some observers attributed the prosecution partly to Senator Charles Grassley’s involvement. Research-fraud prosecutions were described as rare; by comparison, Scott Reuben received six months in prison in 2010 for fabricating data and patients in more than twenty papers.