arXiv caps authors at two paper submissions per month as AI-driven volume doubles

What changed

Starting October 1, 2026, arXiv restricts every submitter to a maximum of two submissions per calendar month, with no more than three submissions active at any given time. The preprint server announced the change on its blog, in a post that surfaced on the Hacker News front page, and described it as a temporary measure while it works out longer-term norms for authors using increasingly capable AI tools and invests in better moderation tooling.

Rate limiting itself is not new at arXiv. According to the announcement, moderators have long had discretion to throttle submitters whose output far exceeded what a person could plausibly produce, and a cap of three simultaneous active submissions has been in place since 2024. What is new is a uniform monthly ceiling applied to all submitters across all subject categories.

The numbers behind the decision

The policy arrives amid a steep climb in submission volume. arXiv reports receiving 40,363 submissions in September 2026, a single-month record, which in turn produced close to 9,000 support tickets for staff and volunteer moderators. Two years earlier, in September 2024, the figure was 20,569, and a decade before that, in September 2016, it was 9,869. In other words, monthly submissions have doubled in just two years.

The growth is unevenly distributed. The cs.AI category has seen submissions rise more than sixfold over the past two years, while other categories have roughly doubled, according to the blog post.

arXiv attributes much of the surge to AI tools that make parts of research and writing faster and easier. Its policy permits AI assistance as long as it is disclosed and the work meets the repository's standards for original, significant research of scholarly interest. But moderators say a growing share of incoming papers fall short: they report more thin papers of narrow scope, more "salami" submissions where one body of work is split into several smaller papers, and a marked rise in dense, AI-written text.

The strain falls on volunteers. Thomas Dietterich, chair of the arXiv Editorial Advisory Council, said a relatively small group of authors submitting large volumes of low-quality work consumes a disproportionate share of moderator attention, delaying papers from authors who follow the guidelines by days or weeks.

How the limits work

The announcement includes a FAQ that clarifies the mechanics:

• Rejections still count. The cap applies to submissions rather than announced papers, because submissions are what consume moderator time. Two submissions in a month, whether accepted or rejected, exhausts that month's allowance.
• The limit spans all categories, so submitting to different subject areas does not create extra capacity.
• Submissions still on hold from a previous month do not count against the next month's allowance, but they do count toward the three-active-submission cap.
• Limits attach to the submitter, not the author list. If one person submits on behalf of several co-authors, only that person's quota is affected, so teams need to coordinate who files each paper.
• A paper deleted before it is announced does not count toward either the monthly or active limits.

arXiv also asks authors with backlogs or conference deadlines to plan their timing so the monthly and active caps do not interfere with their schedules.

Why it matters

arXiv is foundational infrastructure for anyone who works with scientific literature at scale, from individual researchers tracking their field to the software and services that index, summarise and analyse papers. Notably, this policy governs submissions, not reading or retrieval, so it does not directly restrict how the corpus is consumed. But the underlying tension it addresses, machine-accelerated paper production pressing against finite human moderation capacity, is one that every open repository is now navigating.

The move also illustrates a broader shift in quality control. When the practical pace of human authorship acted as a natural brake, discretionary moderation was enough. With that brake gone, arXiv is resorting to blunt numerical limits, which will also constrain legitimate, highly productive authors and complicate conference-season logistics for some teams.

arXiv says it will monitor the effects of the policy on authors, readers and the corpus, and adjust it as needed, leaving open the question of what a durable moderation model for the AI era looks like.