AI news ·
AI-hallucinated citations flood scientific literature as fabrication rates rise 12-fold since 2023
AI-generated fake citations rose more than 12 times between 2023 and 2025, with nearly 147,000 hallucinated references found across major scientific repositories last year. Clinical trials and systematic reviews were among the affected studies.
Fake Citations Generated by AI Are Flooding Scientific Literature
Fabricated references created by artificial intelligence tools have climbed steeply since 2023, according to an analysis of 97 million citations published in Nature. The rate of fake citations in 2025 was more than 12 times higher than in 2023.
A separate audit of major scientific repositories found an estimated 146,900 hallucinated citations in research papers across arXiv, bioRxiv, SSRN, and PubMed Central in 2025 alone. The problem included 28 clinical-trial studies and 79 systematic reviews.
A Widespread Pattern, Not Isolated Cases
The fake citations are not concentrated in a few papers or authored by a handful of researchers. Instead, they appear scattered across many papers, each containing a small number of fabricated references. This pattern suggests researchers are using AI tools without verifying the output.
arXiv, the physics and computer science preprint server, has responded by threatening to ban researchers for one year if their submissions contain hallucinated references.
What's at Risk
The researchers behind the audit warn that hallucinations are "steadily infiltrating knowledge production at scale, threatening both its reliability and equity." Without intervention, the impact could extend from scientific discovery to policy and public understanding.
The problem creates a secondary burden: readers and researchers must now sift through contaminated literature to identify what is real. This adds friction to the research process across all fields that rely on peer-reviewed sources.
For professionals in research roles, understanding how generative AI and LLM tools work - and their limitations - has become essential to maintaining research integrity.