Provenance, Not Prohibition: A Framework for AI in Scholarly Publishing

ABSTRACT

Generative AI is now embedded in scholarly workflows, yet publishing policy has often jumped to permission or prohibition without first defining what is being governed. The central risk is not tool use itself, but loss of provenance: readers, reviewers, and editors must be able to trace how claims, citations, analyses, and interpretations were produced and verified. Harm is already measurable, including fabricated references, distorted summaries, hollow reviews, and disclosure policies that are widely ignored. This commentary proposes a practical framework centered on human accountability: separate capability from responsibility, prioritize provenance over polish, avoid unverifiable bans, and pair disclosure with explicit verification. Because AI now enters every stage of review, the unit of governance is not the static manuscript but the feedback loop around it, and the same obligations bind authors, reviewers, and editors alike. Journals should govern outcomes and professional standards rather than tool lists, building a culture of transparent use instead of surveillance.