AI-hallucinated citations are creeping into papers that shape clinical guidelines, researchers warn
Fabricated references in biomedical papers have increased more than twelvefold since 2023, likely linked to language models, according to a Columbia University-led audit.

A recent audit of 2.5 million biomedical papers has uncovered a disturbing trend: the rate of fabricated references has increased more than twelvefold since 2023. The researchers behind the audit, which involved Columbia University and other institutions, suspect a connection between this surge and the widespread adoption of language models. These AI-generated references are particularly problematic because they often match the topic of the paper, follow proper formatting, and are nearly impossible to detect.
The audit's findings are concerning, given that these papers often shape clinical guidelines. The researchers warn that the infiltration of fabricated references threatens the integrity of medical research and, by extension, patient care. What's more, the vast majority of affected papers – 98 percent, to be exact – have received no response from their publishers, suggesting a lack of scrutiny or oversight.
The emergence of language models has undoubtedly changed the research landscape. While these tools can assist with writing and organization, they can also fabricate references with ease. The resulting citations are often convincing and go undetected by current peer-review processes.
This vulnerability highlights the need for more sophisticated methods to verify the accuracy of references. The researchers' findings serve as a call to action for the academic community, publishers, and funding agencies. They must work together to develop more robust systems for detecting fabricated references and to promote responsible use of language models in research.
Ultimately, ensuring the accuracy and reliability of medical research is crucial for advancing healthcare and patient outcomes. The issue of AI-hallucinated citations underscores the importance of ongoing vigilance in maintaining the integrity of scientific literature. As the use of language models becomes more prevalent, it is essential to address this challenge proactively and collaboratively.
Source: The Decoder