Fabricated Citations Polluting Scientific Literature as AI Hallucinations Spread Through Research Papers
A new study published in the Lancet reveals that fabricated citations—references to non-existent papers—are proliferating across scientific literature, raising concerns about the integrity of academic research. Columbia University researchers analyzing the phenomenon link the trend directly to generative AI tools, which they say are increasingly generating plausible but entirely fictional source material. The findings suggest that a foundational pillar of scholarly verification—citation chains tracing the lineage of ideas—may be corroding at an alarming rate.
The researchers examined citation patterns and identified instances where references pointed to papers that simply do not exist. Rather than errors of omission, these appear to be confidently fabricated citations that mimic legitimate academic footnoting. AI language models, which can produce fluent text that sounds authoritative without grounding in factual sources, are the primary suspects. The phenomenon mirrors what developers have termed "hallucinations"—confident outputs that lack real-world referents. For peer reviewers, editors, and readers, detecting these fabricated references requires investigative effort that existing quality controls are not equipped to handle at scale.
The implications extend beyond individual papers. Citations form the evidentiary backbone of meta-analyses, clinical guidelines, and policy decisions. When that backbone contains phantom references, downstream research built upon it inherits the contamination. Academic publishers and institutions face mounting pressure to develop detection mechanisms and new verification protocols. The study stops short of quantifying the total scope of the problem but signals that it is large enough to warrant systemic intervention. As AI-generated content becomes cheaper and more accessible, the risk of fabricated citations becoming normalized in scientific publishing continues to escalate.