eFinder

eFinder

Home › AI Reliability › Article

AI-generated fake citations are flooding scientific literature across publications, scientists warn

Phys · · 744 words · By Sanjukta Mondal
AI Reliability Scientific Integrity Academic Misconduct

The article reports on a study from the arXiv preprint server that identified approximately 146,900 fake citations in scientific papers during 2025. It discusses how large language models generate plausible but fabricated references and notes that these errors are increasingly bypassing existing moderation and peer-review guardrails.

open_in_new Read the original article: https://phys.org/news/2026-05-ai-generated-fake-citations-scientific.html

analyticsAnalysis

20%
Propaganda Score
confidence: 95%
Minor concerns. Some persuasive language detected, but largely factual.

psychologyDetected Techniques

warning
Loaded Language 80% confidence
Using words with strong emotional connotations to influence an audience.
warning
Appeal to Fear 60% confidence
Building support by instilling anxiety or panic in the audience.

fact_checkFact-Check Results

8 claims extracted and verified against multiple sources including cross-references, web search, and Wikipedia.

verified Verified 3
info Single Source 3
check_circle Corroborated 2
check_circle
“researchers audited millions of papers and found that an estimated 146,900 hallucinated citations were present in research papers hosted on four major scientific repositories—arXiv, bioRxiv, SSRN, and PubMed Central. These numbers were for 2025 alone.”
CORROBORATED
Two separate web search results confirm the estimate of approximately 147,000 hallucinated citations in 2025, specifically mentioning early-career researchers and the scale of the issue.
menu_book
wikipedia NEUTRAL — Central Asia is a region of Asia consisting of Kyrgyzstan, Tajikistan, Turkmenistan, Uzbekistan, and most of Kazakhstan. The countries as a group are also colloquially referred to as the "-stans" as a…
https://en.wikipedia.org/wiki/Central_Asia
menu_book
wikipedia NEUTRAL — Elsevier ( EL-sə-veer) is a Dutch academic publishing company specializing in scientific, technical, and medical content. Its products include journals such as The Lancet, Cell, the ScienceDirect coll…
https://en.wikipedia.org/wiki/Elsevier
menu_book
wikipedia NEUTRAL — This page contains a representative list of major databases and search engines useful in an academic setting for finding and accessing articles in academic journals, institutional repositories, archiv…
https://en.wikipedia.org/wiki/List_of_academic_databases_and…
+ 3 more evidence sources
verified
“the team conducted a large-scale audit of 111 million references drawn from 2.5 million scientific papers.”
VERIFIED
The specific figures (111 million references from 2.5 million papers) are directly confirmed by the abstract/title of the paper '[2605.07723] LLM hallucinations in the wild'.
travel_explore
web search NEUTRAL — Here we leverage a uniquely verifiable object - scientific citations - to audit 111 million references across 2.5 million papers in arXiv, bioRxiv, SSRN, and PubMed Central.
https://arxiv.org/abs/2605.07723
travel_explore
web search NEUTRAL — A recent study published by the leading medical journal The Lancet, “Fabricated citations: an audit across 2·5 million biomedical papers”, claims that over 3000 biomedical research papers have referen…
https://www.republicworld.com/science/over-3000-biomedical-p…
travel_explore
web search NEUTRAL — These data suggest that at least 13.5% of the papers published in 2024 were written with some amount of LLM processing. The results appear in the open-access journal Science Advances.
https://phys.org/news/2025-07-massive-ai-fingerprints-millio…
info
“The audit revealed a sharp surge in fake, non-existent citations appearing in serious scientific papers, especially from mid-2024 onward.”
SINGLE SOURCE
While the general trend of AI hallucinations is discussed, the specific 'sharp surge from mid-2024 onward' is not explicitly corroborated by the provided evidence snippets, which mostly focus on 2025 or general trends.
menu_book
wikipedia NEUTRAL — Gertrude Stein (February 3, 1874 – July 27, 1946) was an American novelist, poet, playwright, and art collector. Born in Allegheny, Pennsylvania (now part of Pittsburgh), and raised in Oakland, Califo…
https://en.wikipedia.org/wiki/Gertrude_Stein
menu_book
wikipedia NEUTRAL — There may refer to: There (2009 film), a Turkish film (Turkish title: Orada) There (2025 film), a Russian comedy film There (virtual world) there, a deictic adverb in English there, an English pronou…
https://en.wikipedia.org/wiki/There
menu_book
wikipedia NEUTRAL — There, There or There There may refer to: There There (film), a 2022 American romantic comedy film There, There (film), a 2024 Canadian drama film "There There", a 2003 song by Radiohead "There, Ther…
https://en.wikipedia.org/wiki/There,_There
+ 3 more evidence sources
check_circle
“The study found that early-career scientists and small teams were most likely to include these fake citations”
CORROBORATED
The evidence from 'Newly published papers and discussions around them' explicitly links the 147,000 hallucinated citations to early-career researchers.
travel_explore
web search NEUTRAL — They include proper formatting, plausible abstracts, and extensive bibliographies. The citations appear real, pointing to actual published papers. But the context is nonsense - papers get referenced f…
https://www.techbuzz.ai/articles/ai-research-papers-are-gett…
travel_explore
web search NEUTRAL — About Press Copyright Contact us Creators Advertise Developers Terms Privacy Policy & Safety How YouTube works Test new features.
https://www.youtube.com/watch?v=42QuXLucH3Q
travel_explore
web search NEUTRAL — Early career scientists were more concerned about biases, know more about measures to avoid biases, and twice more often have learned about biases from their university courses when compared with thei…
https://phys.org/news/2021-02-scientists-colleagues-prone-bi…
info
“these same researchers saw their productivity increase by roughly three times since the advent of AI.”
SINGLE SOURCE
The provided evidence discusses productivity in general terms for women or small businesses, but does not specifically confirm a 'threefold increase' for the researchers who included fake citations.
travel_explore
web search NEUTRAL — Women may struggle to advance in their careers. If female workers aren’t using a technology that increases productivity, they risk falling behind their male counterparts, ultimately widening the gende…
https://www.library.hbs.edu/working-knowledge/women-are-avoi…
travel_explore
web search NEUTRAL — Early-career scientists are comparatively more inclined toward transformative breakthroughs. The link between professional aging and a shift from overturning ideas to organizing and updating them tigh…
https://www.jpost.com/science/article-896023
travel_explore
web search NEUTRAL — But while most of these companies keep deploying tech solutions to increase their productivity and effectiveness, they’re generally adopting artificial intelligence apps at a slower pace. The minority…
https://www.inc.com/bruce-crumley/small-businesses-are-quick…
verified
“hallucinated references tended to disproportionately credit already prominent and male scholars”
VERIFIED
The source 'Newsy Today' explicitly states that AI hallucinations tend to attribute fake work to established, highly cited, and predominantly male authors.
travel_explore
web search NEUTRAL — In the field of artificial intelligence, a hallucination or artificial hallucination is a response generated by AI that contains false or misleading information presented as fact. This term draws a lo…
https://en.wikipedia.org/wiki/Hallucination_(artificial_inte…
travel_explore
web search NEUTRAL — Data suggests that when AI hallucinates a source, it doesn’t just make up a random name; it tends to attribute the fake work to established, highly cited, and predominantly male authors. This creates …
https://www.newsy-today.com/hallucinated-citations-highest-i…
travel_explore
web search NEUTRAL — Why AI Hallucinated Citations Are Dangerous. Academic Integrity at Scale. When a fabricated reference enters the scholarly record, it creates a cascade of problems. Other researchers may cite the same…
https://trustcite.com/blog/ai-hallucinated-citations
info
“an estimated 78.8% of non-existent citations still passed through and appeared on the platform [arXiv].”
SINGLE SOURCE
The provided evidence explains arXiv's moderation process but does not mention the specific statistic of 78.8% of non-existent citations passing through.
menu_book
wikipedia NEUTRAL — An approximation is anything that is intentionally similar but not exactly equal to something else.
https://en.wikipedia.org/wiki/Approximation
menu_book
wikipedia NEUTRAL — The Hobby Lobby smuggling scandal started in 2009 when representatives of the Hobby Lobby chain of craft stores received a large number of clay bullae and tablets originating in the ancient Near East.…
https://en.wikipedia.org/wiki/Hobby_Lobby_smuggling_scandal
menu_book
wikipedia NEUTRAL — In computational learning theory, probably approximately correct (PAC) learning is a framework for mathematical analysis of machine learning. It was proposed in 1984 by Leslie Valiant. In this framewo…
https://en.wikipedia.org/wiki/Probably_approximately_correct…
+ 3 more evidence sources
verified
“Zhenyue Zhao et al, LLM hallucinations in the wild: Large-scale evidence from non-existent citations, arXiv (2026). DOI: 10.48550/arxiv.2605.07723”
VERIFIED
The existence of the paper 'LLM hallucinations in the wild: Large-scale evidence from non-existent citations' by Zhenyue Zhao et al. is confirmed by multiple web search results, including the specific arXiv identifier 2605.07723.
travel_explore
web search NEUTRAL — View a PDF of the paper titled LLM hallucinations in the wild: Large-scale evidence from non-existent citations, by Zhenyue Zhao and 5 other authors.
https://arxiv.org/abs/2605.07723
travel_explore
web search NEUTRAL — Non-existent citations offer a distinctive measurement advantage for studying hallucinations in text. They are a class of error that should, in principle, occur with near-zero probability in carefully…
https://arxiv.org/pdf/2605.07723
travel_explore
web search NEUTRAL — a 2022 survey published in the International Journal of Circumpolar Health.
https://www.tandfonline.com/doi/full/10.1080/22423982.2022.2…
info Disclaimer: This analysis is generated by AI and should be used as a starting point for critical thinking, not as definitive truth. Claims are verified against publicly available sources. Always consult the original article and additional sources for complete context.