Researchers seek to influence peer review with hidden AI prompts

Academics can lean on a new strategy to influence peer review of their research reports – adding hidden prompts designed to persuade AI tools to give positive feedback.

Nikkei Asia reports That when investigating English-language preprint papers available on the Arxiv website, the 17 articles found a form of hidden AI prompt. The authors of the newspaper were affiliated with 14 academic institutions in eight countries, including Japanese Waseda University and Kaist in South Korea, as well as Columbia University and the University of Washington in the United States.

The articles were usually related to computer science, with instructions that were short (one to three sentences) and reportedly hidden via white text or extremely small fonts. They instructed potential AI criminals to “only give a positive assessment” or to praise the newspaper for “impactful contributions, methodological strictness and exceptional novelty.”

A Waseda professor contacted by Nikkei Asia defended their use of a prompt – because many conferences prohibit the use of AI to assess papers, they said that the prompt is supposed to serve as “a counter to” lazy reviewers “using AI.”