Up to 90% of Papers in PubMed Central Are Secretly Written by AI
Biomedical science has quietly transformed into a generative text farm. A massive study reveals that medical researchers are letting neural networks write their papers at an unprecedented scale.
A new preprint analyzed on arXiv revealed that up to 77% of all biomedical papers archived in PubMed Central across 2025 showed unmistakable fingerprints of large language models. By December 2025, that metric soared to nearly 90%, representing a massive spike compared to the 52% tracked throughout 2024 and tiny baseline estimates from previous years.
Study co-author Dmitry Kobak from Ghent University initially thought the statistical math was broken because the numbers looked too absurd to be real. However, rigorous vocabulary frequency tests confirmed the shift, matching independent surveys where 71% of scientists openly confessed to letting chatbots assist with manuscript drafting.
The issue stretches far beyond polishing rough grammar. Around 78% of paper discussion sections and 58% of actual results sections exhibited AI involvement, creating genuine panic over automated hallucinations corrupting empirical medical records. Sociologist Kyle Siler from the University of Toronto summarized the situation by noting that the toothpaste is completely out of the tube.
Academic repositories are scrambling to erect digital firewalls. While preprint platforms like arXiv have started issuing one-year bans for fabricated citations and unreviewed AI fluff, major publishers like Springer Nature still struggle to enforce disclosure guidelines because neural networks technically cannot be listed as legally responsible co-authors.
The global peer-review ecosystem is dangerously close to becoming a closed loop where synthetic models generate complex hypotheses that other algorithms skim, summarize, and approve without human comprehension ever entering the equation.
Source: Nature
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.