AI hallucinations are part of a reproducibility crisis, threatening medicine, law, and journalism as unverified AI output risks entering the permanent record.

Publishers show verification gaps: journals differ in checking references and screening for fake studies, and audits found that 98.4% of papers with fake references were not retracted at the time of review.

Fabricated references undermine the evidence chain for medical guidelines and patient care, because errors at the start can cascade through trials, reviews, and guidelines.

Columbia associate professor Maxim Topaz recalls an AI-generated fabricated reference in a manuscript, illustrating the real risk of AI hallucinations in research.

Beyond academia, misattributed quotes and AI-assisted practices span journalism, law, medicine, and even Nobel-level authors, showing widespread exposure to AI errors.

Topaz calls for embedding verification into research workflows rather than banning AI, warning that delays make cleanup harder as AI apps proliferate.

AI hallucinations occur when models optimize for word patterns over accuracy, and experts—not just students—are susceptible.

A Lancet audit of nearly 2.5 million biomedical papers and 97 million PubMed citations found over 4,000 fabricated references across nearly 3,000 papers, with a sharp rise in 2024 as AI tools spread.

Fabricated references surged more than 12-fold from 2023 to early 2026, moving from about one in 2,828 papers to one in 277 in early 2026.