Governing Generative AI in Peer Review: An Author-Authorised, Version-Synchronised Preprint Pathway

Generative artificial intelligence (AI) in peer review raises questions of permission, human judgment, and fair participation. This normative and conceptual analysis proposes an author-authorised, journal-authorised, version-synchronised public-preprint pathway for bounded external-AI assistance. It aims to broaden opportunities for defensible criticism while maintaining rigorous evaluation and limiting governance to necessary safeguards. Eight cumulative conditions connect voluntary, unanimous author opt-in, journal permission, a rights-cleared public package, round-specific synchronisation, service eligibility, disclosure, human accountability and redress, and exclusion of non-public editorial information. Synchronisation supplies complete, same-version context for checking consequential criticisms without requiring full-manuscript input for every task. An optional open pre-review pilot permits contributions before formal reviewer appointment while preserving invited review. Comparisons separate public access, participation, and AI provision across confidentiality, retention and training, model change, auditability, burden, equity, and complementary questioning. Approved external assistance may support contributors without journal-provided internal-AI accounts; its value depends on voluntary uptake, usable support, contribution quality, and total burdens relative to feasible alternatives. Two-level eligibility checks, proportionate records, staged publication, and independent redress make the proposal inspectable. Effects remain untested. The result is a design for comparative evaluation, not a validated standard, a claim of superiority over internal assistance, or permission to disregard existing policies.