A new analysis reports that AI-writing signatures in biomedical papers have surged dramatically, with the share rising from 52% in 2024 to 77% across 2025, and approaching nearly 90% by December 2025. Researchers analyzed word-frequency patterns characteristic of large language models across biomedical publications archived in PubMed Central. The study is discussed with input from computer scientist Dmitry Kobak at Ghent University, and publication references include an arXiv preprint and work in Science Advances, focusing on indicators of LLM-assisted writing. The findings show discussion sections are most likely to contain AI involvement, reported at 78% versus 58% in results sections. For the biotech sector, the immediate impact is scientific integrity and translational reliability: AI-assisted generation in results sections may carry different risks than in narrative framing. Journals that rely on voluntary disclosure may face increasing difficulty when AI involvement becomes the norm rather than the exception. The analysis points to a workflow problem for peer review—reviewers may need better tooling to assess authenticity when AI-free papers are no longer a reasonable default assumption.