AI Use in Biomedical Papers Reaches 90%, Sparking Concerns Over Fabricated Data

The widespread and accelerating use of artificial intelligence in academic publishing has raised alarms, particularly within the field of biomedicine. A comprehensive study examining data from the PubMed platform, a central repository for medical literature, found that AI assistance was present in 90% of articles published in late 2025. For the entire year, the AI usage rate stood at 77%, underscoring a dramatic and largely unchecked surge in the adoption of generative AI tools by researchers.

“

AI models can sometimes generate unrealistic data ('hallucinations'), which could lead to the infiltration of erroneous medical information into academic literature.

The study, published on the arXiv platform, highlights that AI is not merely being used for basic language corrections but is deeply integrated into the core sections of scientific papers. AI traces were found in 78% of the "discussion" sections and 58% of the "results" sections of the analyzed articles. Dmitry Kobak, a computer scientist at Ghent University in Belgium and a co-author of the study, expressed serious concern about AI's intervention in the results sections. He warned that AI models can sometimes generate "hallucinations", unrealistic data, which could lead to the dissemination of flawed medical information within academic literature.

Furthermore, the use of AI to draft introductory sections may introduce algorithmic biases into academia, potentially compromising the neutrality of scientific discourse. The data suggests that AI adoption is particularly prevalent among researchers whose native language is not English, with experts viewing AI tools as valuable aids for enhancing academic writing and overcoming language barriers. However, the difficulty in distinguishing between simple grammar checks and fully AI-generated text complicates the process for scientific journals to filter for quality content.

“

The toothpaste is out of the tube, it's impossible to put it back in.

"The toothpaste is out of the tube, it's impossible to put it back in," remarked Kyle Siler, a social scientist at the University of Toronto, acknowledging that AI has become an inseparable part of academic life. In response to this uncontrolled proliferation, academic journals worldwide are beginning to mandate that researchers declare their AI usage during the submission process. Yet, experts agree that disclosure alone is insufficient. Lena Holzwarth, a cognitive scientist at the University of Tübingen, emphasized that the maturation of ideas and the process of articulating thoughts are fundamental to scientific methodology. She argues that delegating article writing to AI systems risks diminishing this critical mental effort for scientists.

“

The maturation of ideas and the process of articulating thoughts are indispensable parts of scientific methodology.