AI Generated Content Tests Preprint Moderation

Preprint servers face an uptick in AI assisted submissions and paper mills, testing how they screen quality.

AI Generated Content Tests Preprint Moderation

PsyArXiv removed a July preprint after a reader flagged it for unclear use of AI and missing author affiliation. The work titled Self-Experimental Report Emergence of Generative AI Interfaces in Dream States listed a single author and raised questions about methods and disclosures. PsyArXiv officials said the work violated terms of use due to unclear or incomplete disclosure of AI involvement. Across preprint servers, experts say AI helped submissions are rare but growing. arXiv estimates about 2 percent of submissions are AI generated or from paper mills, while bioRxiv and medRxiv together turn away more than ten manuscripts per day that seem formulaic. A Nature Human Behaviour study shows AI text appears in abstracts across fields; in computer science 22 percent of arXiv abstracts and 10 percent of biology abstracts on bioRxiv show AI generated content, with 14 percent in biomedical journal abstracts. Some researchers say AI can help authors with English language writing, while others warn of misrepresentation when AI is not disclosed.

Moderation faces a clash between openness and quality. Non profit preprint servers must process high volumes while keeping scrutiny light enough to preserve speed. The rise of AI assisted submissions raises questions about how to define authorship, disclosure, and responsibility. It also points to a broader push for transparency in how science is produced. There is a need for scalable checks that do not burden legitimate authors, and for clear standards on when AI involvement should be disclosed. If the system becomes opaque, reader trust may erode, and the open science project could suffer.

Highlights

Clear rules for AI use will help preserve trust in science as AI tools grow more capable