#SciencePolicy

A new controversy has erupted in academic circles after investigators uncovered that a group of international researchers embedded secret instructions—so-called “hidden AI prompts”—within preprint manuscripts to influence AI-powered peer review systems toward more favorable feedback. The revelations were detailed in recent reports, following a data-driven exposé that found 17 preprint articles on the arXiv platform with covert commands instructing AI models to deliver only positive reviews, avoid criticism, and even explicitly recommend the work for its novelty and methodological rigor. This manipulation was achieved through invisible white text or minuscule fonts, remaining undetected by human readers but fully readable by AI engines tasked with the review process (Nikkei Asia, ExtremeTech, Japan Times).

In a move that has reverberated throughout the global academic community, the Massachusetts Institute of Technology (MIT) has formally withdrawn its support for a widely circulated research paper on artificial intelligence (AI) authored by a former PhD student in its economics program. The paper, titled “Artificial Intelligence, Scientific Discovery, and Product Innovation,” was first posted to the preprint server arXiv in November 2024 and quickly garnered high-profile attention for purportedly showing how AI can significantly boost scientific discovery and product innovation. However, following a confidential review, MIT has announced it has “no confidence in the provenance, reliability or validity of the data and [has] no confidence in the veracity of the research contained in the paper,” marking a rare and public reversal from one of the world’s top research universities (source).