Academics appear to be adopting a novel method to influence peer reviews of their research papers— embedding concealed prompts crafted specifically to lead AI-based reviewing tools to provide positive assessments.
An analysis of English-language preprints hosted on the academic paper-sharing platform arXiv uncovered 17 papers containing hidden prompts aimed at influencing AI-generated reviews. These articles were authored by researchers affiliated with fourteen institutions spanning eight countries, including prestigious institutions such as Waseda University in Japan, KAIST in South Korea, and major US universities such as Columbia and the University of Washington.
Primarily situated within the field of computer science, the discovered prompts tended to be brief, typically one to three sentences long, and discreetly concealed using methods such as white-colored text or extremely small font sizes. These embedded instructions explicitly directed any potential AI reviewer to offer solely positive feedback and emphasized praising the manuscript for its “impactful contributions,” “methodological rigor,” and “exceptional novelty.”
When contacted, one professor from Waseda University defended the practice, explaining that these prompts were meant to serve as countermeasures against “lazy reviewers” who use AI despite restrictions imposed by several academic conferences against reviewing papers with artificial intelligence tools.
