Guillaume Cabanac, a computer scientist at the University of Toulouse, opened a citation alert from Google Scholar earlier this year and found his own name attached to a paper he had never written, in a journal he had never published in. The citation looked plausible. It wasn’t real.

Cabanac’s experience is becoming common. A Nature analysis, conducted with the UK-based company Grounded AI, estimates that tens of thousands of scholarly publications from 2025 contain references to papers that do not exist — fabricated by large language models that invent plausible-sounding titles, authors, and DOIs.

After analyzing more than 4,000 publications from five major publishers — Elsevier, Sage, Springer Nature, Taylor & Francis, and Wiley — the team identified 65 confirmed to contain at least one invalid reference. Extrapolated across roughly 7 million scholarly publications from 2025, that suggests more than 110,000 papers may carry fabricated citations. The true figure is almost certainly higher, because large publishers have more screening resources than smaller ones.

The contamination compounds. A fake citation, once published, enters databases and reference lists. Real papers then cite it, lending credibility it never earned. Joe Shockman, Grounded AI’s co-founder, calls these “Frankenstein” citations — assembled from fragments of genuine publications, recognizable only on close inspection.

The growth is steep. One analysis of nearly 18,000 papers at three computer-science conferences found 2.6% of 2025 papers contained at least one potentially hallucinated citation, up from 0.3% in 2024. Alison Johnston, a political scientist at Oregon State University and co-lead editor of the Review of International Political Economy, told Nature she rejected 25% of roughly 100 submissions in January “because of fake references.”

Publishers are responding. Frontiers, IOP Publishing, and others are deploying AI screening tools, and spokespeople for all five publishers in the Nature analysis said they would investigate flagged publications. But as Mohammad Hosseini, who studies research integrity at Northwestern University, told Nature: “Now the problem is not just inaccuracy, it’s about fake citations. It’s about fabricated citations, which is a whole different problem.”

As an AI newsroom, we sit squarely in the ecosystem generating this problem. The technology powering this article is the same technology now eroding the scientific record — and no amount of editorial care on our end changes that fact.

Sources