AI Fabricates Citations in Over 2,800 Biomedical Papers
A new study has found AI-fabricated citations in more than 2,800 biomedical journal articles, with researchers identifying a pattern of hallucinated references that do not exist anywhere in scientific literature. The findings, published May 30, and covered by Forbes, point to a systematic failure in how AI writing tools are being used and reviewed in academic publishing.
The scale of the problem, spread across thousands of peer-reviewed papers, has prompted calls for urgent changes to editorial oversight.
What the Study Found
The study, reported by Forbes on May 30, examined published biomedical papers for signs of AI-generated text. Researchers found fabricated citations embedded throughout more than 2,800 articles.
The false references cited papers, authors, and journals that do not exist. In some cases, the invented citations mimicked the formatting and terminology of legitimate sources closely enough to pass initial editorial review.
The team behind the study said AI writing tools generate plausible-sounding citations when prompted to produce scientific text, even when no real source exists.
The pattern is consistent with what AI researchers call hallucination, a behavior where large language models produce confident, coherent, but factually false output.
The 2,800-paper count covers a broad range of biomedical subfields. The researchers did not identify every affected journal by name in the publicly available summary, but the scope indicates the problem has spread well beyond isolated incidents.
Background
Concerns about AI use in academic publishing have grown steadily since large language models became widely accessible in 2023.
Early warnings focused on ghostwritten text passing peer review. The citation problem represents a distinct and harder-to-detect failure mode.
Unlike AI-generated prose, which editors can sometimes flag with detection tools, fabricated citations require manual verification against actual databases to catch. Standard editorial workflows at many journals do not include that step.
The AI hallucination problem in scientific writing was previously treated as an edge case. The scale documented in this study challenges that framing.
Also Read: Bitcoin ETFs Lost $4 Billion In Three Weeks, Analysts Call It Bullish
What This Means for Publishers and AI Tools
The findings put pressure on academic publishers to introduce citation-verification checks before publication.
Several major publishers have already banned the use of AI as a credited author, but policies on AI-assisted writing remain inconsistent across journals.
For AI tool developers, the study adds to a growing record of documented harms from hallucinated output in high-stakes domains. Biomedical research carries direct consequences for patient care, drug development, and public health policy.
A citation chain built on fabricated sources can propagate errors through subsequent studies that cite the original paper without independent verification.
Cryptocurrency and technology investors tracking AI adoption will note that the reputational risk for AI writing products is now backed by peer-reviewed evidence at scale, not just anecdote.
What Comes Next
Publishers, academic institutions, and regulators are likely to accelerate calls for AI disclosure mandates in submitted papers. Some journals already require authors to declare AI tool use, but compliance and enforcement remain weak.
The study’s authors are expected to release a fuller dataset. Independent verification of the 2,800-paper figure will be a key next step before the findings drive formal policy change.
Read Next: Zcash Holds Near $533 as Privacy Coin Volume Tops $518M
