Purpose: This review article examines, from an editor’s perspective, how generative artificial intelligence (AI) can be integrated into academic writing, peer review, and editorial workflows while protecting scientific soundness.Current concepts: Editors increasingly encounter manuscripts containing fabricated or unverifiable references and receive post-publication error reports, trends that have been amplified by AI-assisted checking and meta-research. An assessment of 100 Korean medical journals (January–March 2024) found that only 18 journals provided explicit guidance on AI use. Among journals with such policies, most prohibited listing AI as an author and emphasized author accountability, while disclosure requirements, permitted uses, and sanctions varied. Reviews of top-ranked medical journals similarly reveal heterogeneous approaches to disclosure and to reviewers’ use of AI under confidentiality constraints. Some journals, such as Journal of Educational Evaluation for Health Professions, make disclosure optional because AI-detection tools are unreliable and AI-assisted and human-written text are often intermingled.Discussion and conclusion: Natural language–based vibe-coding can also enhance peer review when prompts are designed to ensure reproducible analyses. Confidentiality safeguards and options to opt out of model training should be implemented whenever manuscripts or data are uploaded. Because AI-text screening tools produce variable results and the boundary between acceptable assistance and impermissible primary generation remains unclear, detection should not be relied upon as the sole enforcement mechanism. Instead, authors, reviewers, and editors should serve as supervisors responsible for design decisions and interpretation, as well as for ensuring the availability of data and protocols, while preparing for faster, AI-enabled submission-to-publication workflows.
Sun Huh (Tue,) studied this question.