AI audits reveal surge in fake citations in biomedical papers

Biomedical research papers are seeing a sharp increase in fabricated citations that do not correspond to real academic works, according to a Lancet study.

This trend threatens the reliability of the scientific record by introducing nonexistent evidence into peer-reviewed literature. Because medical professionals rely on these citations to inform patient care and new drug development, the presence of "hallucinated" data could lead to dangerous clinical misconceptions.

AI auditors examining biomedical literature identified 4,046 fake references ^[1]. These fraudulent citations were found across 2,810 different biomedical papers ^[1]. The data indicates a 12-fold rise in these occurrences since 2023 ^[1].

Researchers said the surge is linked to the use of AI-generated text, which often creates plausible-sounding but entirely fictional references. This phenomenon, known as hallucination, occurs when large language models predict a sequence of words that looks like a citation without verifying its existence in a real database.

Beyond AI errors, the audits pointed to coordinated peer-review fraud as a contributing factor. In these instances, the fabrication of references is used to bypass the scrutiny of the academic community, allowing low-quality or fraudulent research to appear legitimate.

The contamination is appearing in papers worldwide, complicating the efforts of other scientists to build upon existing knowledge. When a researcher attempts to verify a claim using a fake citation, the resulting dead end slows the pace of legitimate discovery and wastes institutional resources.