Lancet study: AI-generated fake citations in medical research increased 12-fold - nearly 3,000 peer-reviewed papers affected
What it really says
A research team from Columbia University School of Nursing, led by Professor Maxim Topaz, used AI to systematically check 2.5 million biomedical publications from PubMed Central (January 2023 to February 2026) for fabricated references. The results, published as a research letter in the prestigious journal The Lancet on May 7, 2026: among 97.1 million verified references, the researchers found 4,046 fake citations across 2,810 peer-reviewed papers - references that simply do not exist. The fabrication rate has increased more than 12-fold since 2023: in 2023, one in 2,828 publications contained at least one fabricated citation; by 2025, it was one in 458; and by early 2026, every 277th publication contains a fake citation. The sharpest increase began in mid-2024, coinciding with the widespread availability of generative AI writing tools. Particularly alarming: 98.4% of affected publications had received no publisher action at the time of the study - no correction, no retraction, no notice.
Our assessment
This study deserves a serious risk classification because it documents a concrete, measurable problem with direct implications for healthcare. When doctors base treatment decisions on studies whose references are fabricated, patient safety is at risk. The problem is not AI tools themselves, but their misuse: generative AI models are known to produce plausible-sounding but nonexistent references - so-called hallucinations. When researchers incorporate these unchecked into their publications, it undermines the foundation of evidence-based medicine. The 12-fold increase since 2023 shows the problem is growing exponentially. On the positive side, the Columbia researchers also used AI to uncover the fabrications - an example of how AI can be both part of the problem and part of the solution. Automated verification procedures at publishers before studies are published are urgently needed.
Relevance for Germany
For Germany, this study is relevant for several reasons. German doctors and hospitals base their treatment guidelines on international biomedical research - if fabricated sources are infiltrating that literature, it affects German patient care as well. The German Research Foundation (DFG) published guidelines on good scientific practice with AI in 2024, but implementation in practice is patchy. Additionally, according to a Bitkom study, 58% of Germans use AI tools - if even scientists are uncritically adopting AI-generated misinformation, the question arises how well laypeople can assess the quality of AI outputs. The study underscores the need for media literacy and critical thinking when dealing with AI-generated content - a topic that also needs to be addressed more strongly in German education policy.
Fact check
The primary source is the research letter published in The Lancet on May 7, 2026, by Maxim Topaz and colleagues from Columbia University School of Nursing. The numbers (2.5 million publications, 97.1 million references, 4,046 fake citations in 2,810 papers) come directly from the Lancet publication. The 12-fold increase in fabrication rate and the 'every 277th publication' figure are consistently reported by Nature, Retraction Watch, CIDRAP, and Columbia University itself. The study gained renewed broad coverage on May 26, 2026, including from Nature and Winbuzzer.
Source
- • https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(26)00603-3/fulltext
- • https://www.nursing.columbia.edu/news/nearly-3-000-peer-reviewed-medical-papers-have-fake-citations-columbia-nursing-ai-assisted-audit-finds
- • https://www.nature.com/articles/d41586-026-00748-w
- • https://retractionwatch.com/2026/05/07/one-in-277-pubmed-indexed-papers-in-2026-shows-fabricated-references-says-analysis/