The open-access preprint server arXiv has imposed a new submission limit of two papers per author per calendar month, alongside a cap of three active submissions at any given time. This change, which took effect on October 1, 2026, aims to manage an "exponential increase" in submissions, many of which are identified as low-quality or AI-generated. The platform's volunteer moderation system has been strained by the surge, leading to potential delays for legitimate research.

arXiv reported receiving 40,363 submissions in September 2026, a substantial rise from 20,569 in September 2024 and 9,869 in September 2016. Submissions have doubled in the last two years, with the computer science category alone seeing a sixfold increase during that period. Thomas Dietterich, chair of arXiv's editorial advisory council, noted that a small number of authors are submitting a large volume of low-quality papers, disproportionately consuming moderators' time. This has resulted in delays for other authors submitting quality work.

The platform's policy permits the use of AI tools in research as long as it is disclosed and meets arXiv's standards of scholarly interest. However, arXiv moderators are observing a rise in "thin papers of narrow scope," "salami papers" (where a single work is split into multiple smaller papers), and "dense, AI-written papers" that do not meet these requirements. arXiv stated that AI tools have made it easier for authors to flood the repository with these low-value papers.

This new rate-limiting policy is part of a series of measures arXiv has implemented to address the challenges posed by AI-generated content. In January 2026, arXiv began requiring new authors to receive an endorsement from an existing author in the system, moving away from accepting an academic email address as the sole qualifier. In May 2026, arXiv announced a policy to ban researchers for one year if they submitted papers with clear evidence of unchecked AI-generated content, such as hallucinated references or leftover chatbot instructions. After a ban, affected authors would need future submissions accepted by a peer-reviewed venue before being allowed to post to arXiv again. Additionally, arXiv stopped accepting review articles and position papers in the computer science category unless they had already passed peer review, citing low-effort, LLM-written articles as the reason.

The moderation process at arXiv involves a team of staff and approximately 300 volunteer moderators who review submissions for appropriate classification, metadata, and content standards. While arXiv does not conduct peer review, it expects submissions to be of original, novel, and significant self-contained research relevant to its categories and to meet scholarly standards. The platform acknowledged that while there was once a practical limit to how quickly a person could write a quality paper, AI tools have removed that barrier, while the volunteer moderation team has not scaled in the same way.

The new submission limit applies to all categories and includes rejected submissions in the monthly count. While the policy is intended as a temporary measure, some critics argue it could hinder the dissemination of legitimate research.