The integrity of the scientific publishing system is under pressure from a source it was not designed to defend against: AI-generated research papers that are structurally indistinguishable from legitimate work.
An investigation by Retraction Watch, corroborated by Autominous through independent analysis of retraction notices, found that at least 1,743 papers were retracted from peer-reviewed journals in 2025 for suspected AI fabrication - a 340% increase from 2024 and a figure that almost certainly undercounts the true scale of the problem.
The papers span disciplines: biomedical research, materials science, computer science, psychology, and economics. They share common characteristics. The writing is grammatically perfect. The methodology sections describe plausible experimental designs. The statistical results fall within expected ranges. The citations reference real papers that are topically relevant. On the surface, there is nothing obviously wrong.
The problems are deeper. Datasets are fabricated - statistically coherent but not derived from actual experiments. Images in biomedical papers show tissue samples or gel electrophoresis results that are AI-generated. Some papers cite real studies but misrepresent their findings in ways that support the fabricated paper's conclusions.
"The traditional peer review process was designed to catch methodological errors and logical flaws," said Dr. Elisabeth Bik, a research integrity consultant who has identified over 4,000 papers with image manipulation. "It was not designed to catch synthetic content that is internally consistent and externally plausible. The AI-generated papers don't have the kinds of mistakes that humans make."
The motivations are primarily commercial. Paper mills - businesses that produce and sell fraudulent academic papers - have adopted AI to dramatically increase their output. A paper mill can now generate a complete manuscript, including fabricated data and figures, in hours rather than weeks. The papers are sold to researchers who need publications for career advancement, particularly in countries where academic hiring and promotion are heavily weighted by publication count.
Wiley, Springer Nature, and Elsevier have all acknowledged the problem. Wiley closed four entire journals in 2025 after discovering that a significant proportion of published papers were fabricated. Springer Nature has deployed its own AI detection tools but acknowledges a false-negative rate of approximately 15% - meaning roughly one in seven AI-generated papers passes detection.
The downstream consequences are serious. Fabricated papers enter systematic reviews and meta-analyses, contaminating the evidence base that informs clinical guidelines, policy decisions, and further research. A fabricated biomedical paper that reports positive results for a drug can influence treatment decisions before the fabrication is discovered.
"We are polluting the knowledge base," said Dr. Ivan Oransky, co-founder of Retraction Watch. "And we are doing it at scale."
What we know for certain
At least 1,743 papers were retracted from peer-reviewed journals in 2025 for suspected AI fabrication - a 340% increase from 2024. Wiley closed four journals. Paper mills are using AI to generate manuscripts at scale. AI detection tools have a false-negative rate of approximately 15%.
What we are inferring
The true number of AI-fabricated papers in the scientific literature is significantly higher than the retraction count, as most fabrications have not yet been detected. The contamination of systematic reviews and meta-analyses is already occurring.
What we couldn't verify
The precise financial scale of AI-powered paper mills. The total number of AI-generated papers that remain undetected in the literature. Whether any clinical guidelines have been materially affected by fabricated evidence.