Fake AI-Generated Citations Flooded Research in 2025
Researchers warn that fake AI-generated citations spread across scientific papers in 2025, raising concerns about academic integrity and AI training data.
A quiet credibility crisis unfolded across academia in 2025.
As researchers increasingly adopted generative AI tools for drafting papers and managing references, fabricated citations began slipping into scientific literature at an unprecedented scale. What initially appeared to be isolated mistakes has now emerged as a growing systemic problem affecting journals, preprints, and academic databases worldwide.
A recent cross-institutional study estimated that nearly 150,000 fake AI-generated references entered research outputs during 2025 alone, raising concerns about the reliability of the scientific record.
The scale of the problem surprised researchers
The study, conducted by researchers affiliated with institutions including Cornell University, University of California, Los Angeles, and University of California, Berkeley, analysed roughly 111 million citations across 2.5 million academic papers hosted on repositories such as arXiv, bioRxiv, SSRN, and PubMed Central.
Researchers identified references whose titles could not be verified through major academic databases. According to the analysis, the sharpest increase in fabricated citations began around mid-2024, aligning with the rapid mainstream adoption of AI writing assistants capable of generating references automatically.
The findings also revealed how easily fake citations travelled through publishing systems. Nearly 79% of fabricated references reportedly passed arXiv screening processes, while around 85% of fake citations appearing in bioRxiv preprints later survived into peer-reviewed journal versions indexed in PubMed Central.
What exactly is a hallucinated citation?
A hallucinated citation is an AI-generated reference that appears legitimate but does not actually exist. These fake citations may include invented paper titles, fabricated Digital Object Identifiers (DOIs), incorrect journal names, or mismatched author and publication details.
Because they are formatted in convincing academic language, they can easily escape notice during rushed writing or peer-review workflows. In many cases, researchers may not intentionally submit false references. Instead, authors using AI assistants sometimes assume the generated citations are accurate without independently verifying them first.
That creates a larger risk once these papers become publicly available and get indexed by academic databases.
Why the issue could become long-term
Researchers warn that the problem extends beyond isolated publication errors. Modern AI systems are increasingly trained on massive collections of publicly available academic material. If fabricated references remain embedded inside those datasets, future AI models may repeatedly reproduce and amplify the same fake citations.
This creates a dangerous feedback loop where false information gradually becomes part of the training infrastructure powering future research tools. The risks appear especially serious in biomedical research.
Separate audits cited in the same coverage identified more than 4,000 fabricated references across 2,810 peer-reviewed biomedical papers, with the number of affected studies continuing to rise through early 2026.
Researchers warned that clinical reviews, healthcare recommendations, and evidence-based guidance could eventually be compromised if fabricated references remain unchecked.
Why fake citations spread so quickly
Several pressures combined to accelerate the issue. Academic publishing systems are already stretched by growing submission volumes and limited reviewer capacity. At the same time, preprint platforms prioritise rapid publication, allowing some low-visibility errors to remain undetected before papers move into journals.
Meanwhile, AI writing tools increasingly automate citation generation as part of drafting workflows, encouraging researchers to work faster while reducing manual verification. The combination created an environment where plausible-looking but non-existent references could blend into otherwise legitimate research papers.
The growing push for safeguards
Researchers are now calling for stronger verification systems across academic publishing. Suggested measures include automated reference checks against databases like Crossref and Semantic Scholar, mandatory machine-readable citation identifiers, and stricter editorial screening during submission.
Some experts also argue that universities and research institutions should introduce clearer policies around AI-assisted writing and require authors to manually verify at least a portion of citations before publication.
The surge of fake citations in 2025 may ultimately become a warning for the research community. AI can accelerate scholarship dramatically, but without stronger safeguards, speed could come at the cost of scientific trust.