AI Paper-Review Arms Race Emerges in Scholarly Publishing

Summary

This paper argues that AI-assisted research production and AI-mediated paper evaluation should be studied as an interacting system rather than as separate developments. The authors synthesize 230 scholarly publications and institutional records through six connected dynamics: production scaling, evaluation automation, evaluation manipulation, defensive mechanisms and policy responses, evasion and side effects, and long-horizon ecosystem feedback. The reviewed literature indicates a progression in which cheaper and faster research production increases pressure on scholarly evaluation, while AI makes evaluation more scalable and repeatable. Because evaluators may exhibit regularities, participants can exploit them, prompting institutions to introduce technical safeguards and policy controls. Those responses can produce further evasion, redistribute errors and workload, and influence the scholarly records later reused by research and evaluation systems. Evidence is strongest for large-scale production and evaluation, reproducible manipulation, and institutional responses. The authors say that post-policy adaptation and long-horizon feedback at the level of individual scholarly artifacts remain less directly observed. They therefore frame scholarly publishing as an evolving interaction between human participants and AI systems, rather than as a collection of isolated AI capabilities.