Regulating the unseen hand: AI, authorship, and trust in medical science

Abstract

Large language models (LLMs) have transformed medical research and scientific publishing by facilitating manuscript preparation, literature synthesis, and editorial processes, yet pose significant threats to research integrity through generation of potential pseudoscientific content. Current AI detection algorithms demonstrate inconsistent reliability, particularly against paraphrased or humanized content, while LLM integration in peer review compromises expert critical evaluation and homogenizes scientific discourse. These systems exhibit documented bias against non-male, non-white researchers, compounding ethical concerns. Heterogeneous editorial policies regarding AI disclosure across medical journals create regulatory gaps enabling undetected misconduct. However, excessive focus on detection over content quality risks establishing counterproductive "AI phobia" that impedes legitimate technological integration. Preserving research credibility requires standardized disclosure frameworks, enhanced detection algorithms, comprehensive privacy safeguards, and mandatory AI watermarking systems to maintain scientific integrity while accommodating technological advancement in research practices.