BENGALURU: Hallucinated citations in scientific papers have exploded since the rise of AI tools. At least 1,46,932 fabricated references generated by artificial intelligence entered the scientific record in 2025 alone, and the vast majority of those detected in preprints survived peer review into journal articles.That is the central finding of a large-scale study by researchers at Cornell University, UCLA and UC Berkeley, who analysed 111 million citations across 2.5 million research papers published between 2020 and 2025 on arXiv, bioRxiv, SSRN and PubMed Central.The study, titled “LLM hallucinations in the wild”, tracked citations whose titles could not be verified against major academic databases including Semantic Scholar, OpenAlex and Google Scholar. By comparing post-2022 trends against pre-ChatGPT error baselines, researchers isolated the likely contribution of AI-generated hallucinations to the surge.The numbers are stark. By Aug 2025, hallucinated citation rates had climbed to nearly 2% in SSRN papers, 0.4% in arXiv, 0.3% in PubMed Central and 0.2% in bioRxiv, with monthly fake citation estimates reaching 8,140 in PubMed Central alone.The steepest rise began around mid-2024, roughly 18 months after ChatGPT’s public release, as AI tools evolved from writing assistants into citation-generation engines.The contamination is not concentrated in obviously fraudulent papers. Researchers found that fake references are typically sprinkled sparsely across otherwise legitimate manuscripts, suggesting many researchers are copying AI-generated citations without verifying them.The problem skews toward specific demographics. Authors of hallucinated citations tended to be less experienced, but their output is growing fast, rising 3.13 times faster on SSRN and more than doubling on bioRxiv compared to matched peers by 2025.Solo researchers and small teams were also disproportionately represented. When hallucinated references pointed to real scientists, they disproportionately favoured prominent scholars — those cited had 68.8% more prior publications and 58.3% more citations than average.Existing safeguards are failing. Nearly 78.8% of fake citations passed arXiv moderation, and among bioRxiv preprints later published in PubMed Central-indexed journals, 85.3% of hallucinated references made it into the final published versions.The researchers warn the problem may now be self-reinforcing: as fabricated references embed themselves in open-access repositories and citation databases, future AI models trained on that corpus risk absorbing — and reproducing — the same hallucinations.Study In Lancet Too WarnsIn a separate study titled “Fabricated citations: an audit across 2·5 million biomedical papers” published in the The Lancet, researchers found a sharp rise in fabricated citations in biomedical research papers.The study, conducted by researchers from Columbia University and other institutions, analysed biomedical papers published between 2023 and early 2026. It found more than 4,000 fabricated references embedded across 2,810 peer-reviewed papers.The audit found the rate of fabricated references rose dramatically over the three-year period. In 2023, roughly one in 2,828 papers contained at least one fabricated citation. By 2025, the figure had worsened to one in 458 papers, and by early 2026, it had climbed further to one in 277 papers.One of the most striking examples cited in the study involved a 2025 paper in an open-access oncology journal on ureteroileal surgical techniques. Researchers found that 18 of the paper’s 30 verified references — or 60% — were fabricated.The authors linked the surge partly to widespread adoption of LLMs which are known to “hallucinate” fake citations.Warning that fabricated citations could compromise clinical guidelines and systematic reviews, the researchers urged publishers to introduce automated reference verification systems before papers are accepted for publication. Nearly 98% of the affected papers had not faced any publisher action at the time of the audit, the study noted.
