Artificial intelligence may be accelerating a growing research integrity problem within the biomedical literature. A new analysis published in The Lancet found that fabricated references—citations pointing to papers that do not exist—are appearing in scientific publications at a rapidly increasing rate.
Researchers led by Maxim Topaz analyzed nearly 2.5 million PubMed-indexed papers and discovered that approximately one in every 277 papers published during the first seven weeks of 2026 contained at least one fabricated citation. That represents a sharp increase from one in 458 papers in 2025 and one in 2,828 papers in 2023.
The team reviewed more than 97 million references and identified 4,406 fabricated citations appearing across 2,810 papers. Researchers used AI-assisted methods to distinguish nonexistent references from formatting inconsistencies and abbreviated titles. According to the study, the steepest increase in fabricated citations began in mid-2024, roughly coinciding with the widespread adoption of generative AI writing tools.
For laboratory professionals, pathologists, and clinical researchers who rely heavily on published evidence to support testing decisions, validation studies, and operational planning, the findings raise questions about the reliability of scientific literature in an era of increasingly automated content generation.
Debate grows over corrections and retractions
Despite the scope of the problem, the researchers found that more than 98% of papers containing fabricated references had not received any correction, expression of concern, or retraction at the time of the audit.
Publishers are beginning to respond. Some organizations are investing in technology and editorial review processes designed to identify questionable citations before publication, while others are exploring systemwide reference-screening programs.
Not everyone agrees, however, on how journals should handle papers containing fabricated references.
Some research integrity experts quoted in the article argue that authors bear responsibility for every citation in a manuscript and that papers containing nonexistent references should be retracted. Others contend that the response should depend on whether the fabricated citation materially affects a study’s conclusions.
Topaz noted that roughly 91% of the affected papers contained only one or two fabricated references, suggesting many cases may result from researchers using AI-generated content without independently verifying citations. In other instances, however, fabricated references appeared more extensively throughout manuscripts and could potentially influence scientific claims.
Experts caution that fabricated references may represent only the most visible manifestation of a broader challenge. AI-generated citations can also be inaccurate, incomplete, or biased while still pointing to real publications, making them considerably harder to identify and correct.
The study’s authors recommend automated reference verification during manuscript submission, expanded integrity tracking within indexing databases, and retrospective reviews of published literature. They also expressed concern about review articles, which showed a fabrication rate 57% higher than other paper types.
As AI tools become increasingly embedded in scientific writing workflows, the findings suggest laboratories, publishers, and researchers may need stronger safeguards to ensure the literature underpinning clinical and scientific decision-making remains trustworthy.
On the same topic, in October 2025, The Dark Report published an article on a study suggesting scientific fraud is more widespread than many researchers believe, fueled by organized “paper mills” that sell or place fabricated manuscripts in academic journals. Researchers found evidence of coordinated networks of authors and editors helping fraudulent studies pass peer review, raising concerns about the reliability of published research and the growing pressure on scientists to publish.
This article was created with the assistance of Generative AI and has undergone editorial review before publishing.
