Biomedical science is facing a crisis of systemic contamination: AI hallucinations have begun infiltrating peer-reviewed literature en masse, poisoning the evidence base. According to a report by a team led by Maxim Topaz from Columbia University, published in The Lancet, the number of fabricated citations in biomedical papers has skyrocketed more than twelvefold since 2023.
An audit of 2.47 million papers from the PubMed Central archive reveals a terrifying trajectory. While the rate held steady at 4 fakes per 10,000 articles throughout 2023, the figure jumped to 56.9 in the first weeks of 2026. These hallucinations are dangerously persuasive—they employ correct formatting and real researcher names to support arguments that are seamlessly woven into the study's context.
Key Takeaways: The Scale of the Threat to Evidence-Based Medicine
The number of fake citations in scientific works has increased more than 12 times. Review articles show a fabrication rate 57% higher than primary research. Out of 97.1 million verified references, 4,046 were confirmed forgeries spread across 2,810 articles. 98.4% of compromised papers have yet to be retracted or flagged by publishers.
For the MedTech and pharmaceutical sectors, this trend is more than an academic curiosity; it is a direct legal liability. As large language models like ChatGPT have become standard tools for manuscript preparation, the risk of implementing flawed treatment protocols has become critical. This is not a technical glitch, but a rupture in the chain of evidence.
In one analyzed urology paper, 18 out of 30 references were pure fiction, yet they appeared so plausible they easily slipped through the net of traditional peer review.
Industry Response and Consequences
Industry reaction has been sluggish and largely ineffective. Despite the identification of thousands of fake citations, the vast majority of these works remain accessible without any warning labels. While platforms like Arxiv are introducing sanctions, the scientific community is demanding mandatory automated citation verification.
Without a retrospective purge of archives and the implementation of integrity-check metadata, the clinical guidelines that govern patient lives will remain a shaky foundation built on neural network hallucinations. This issue demands immediate regulatory intervention and a fundamental shift in scientific publishing standards.