arXiv began limiting researchers to two submissions per month on Thursday, Oct. 1, as the preprint repository reported record submission volume and growing pressure on its moderators. The new cap is meant to spread review work more evenly, but arXiv’s figures do not show how much of the increase consists of AI-generated or low-quality papers.
The new policy allows each submitter to send no more than two papers per calendar month across all categories, with a maximum of three active submissions at a time. A rejected paper still counts toward the monthly allowance.
The limit applies to the person who submits a paper, not separately to every co-author. Collaborators can submit under their own accounts, but arXiv says they should coordinate. Researchers with a backlog or a deadline may need to plan when to send work.
How does the arXiv submission cap work?
The three-active-submission ceiling is not new: arXiv says it dates to 2024. The new restriction is the fixed monthly limit. Before it took effect, rate limits were applied mainly at moderators’ discretion, according to the repository.
ArXiv describes the change as temporary, while it considers how to handle submissions in an era of increasingly capable AI tools and works on moderation processes. The cap controls how many papers a submitter can send; it does not introduce a new test of scholarly quality.
The repository’s existing standards call for work to be original, novel, significant, self-contained and relevant to its category. In a statement reported by 404 Media, arXiv said: “This policy update strives to fairly distribute the incredibly valuable work of our volunteer moderators who check that submissions meet these standards”.
Thomas Dietterich, chair of arXiv’s Editorial Advisory Council, said a small proportion of authors submit many low-quality papers, taking a disproportionate share of moderator time and potentially slowing other papers. That is the repository’s rationale for changing the rate limit, not a measured assessment of the cap’s effects.
What do arXiv’s submission figures show?
ArXiv counted 40,363 submissions in September 2026, a record for the repository. Its totals were 20,569 in September 2024 and 9,869 in September 2016. The recent figures show a sharp rise in volume, but they do not identify how papers were written or evaluate their research quality.
ArXiv says the number of submissions to its cs.AI category grew more than sixfold in the two years through September 2026, while other categories doubled. It also said the September volume generated almost 9,000 support tickets for staff and moderators. Those figures describe activity and workload; they do not independently establish that AI caused the increase or that the submissions were poor.
How much of the rise is linked to AI?
ArXiv says moderators have seen more papers that are narrow in scope, fragmented or written with AI. The repository has not published an independently audited count showing what share of submissions use AI or meet its definition of low quality.
Free newsletter
Get the morning briefing
Start each day with the AI stories that matter and why — a short, free email from our newsroom.
In a January report, Science quoted Ralph Wijers, arXiv’s editorial council chair, saying that a large fraction of submissions rejected by the repository were AI-generated. Wijers estimated that arXiv’s rejection rate had risen from about 4% to 10%–12%. Those are his assessments, not results from an independent audit of submissions.
The distinction matters: a rising total can show that moderators face more material to handle, but it cannot establish why submissions increased or how many fail the repository’s standards. ArXiv’s reported observations may identify a real moderation concern, while leaving its scale and connection to AI unquantified.
The submission cap follows other steps arXiv has taken. On Oct. 31, 2025, it announced stricter rules for computer-science review and position papers that had not passed peer review. On Jan. 21, 2026, an endorsement requirement for first-time submitters took effect, Science reported. On May 19, Nature reported that arXiv would bar authors for a year in cases involving hallucinated references or other conclusive signs that AI-generated material had not been checked.
What could the cap mean for researchers?
For a researcher with more than two ready papers in a month, the limit can mean choosing which one to submit first. Because rejected submissions count, a paper that does not pass moderation can also use a slot that the submitter might otherwise have used for another manuscript. ArXiv advises researchers with backlogs or conference timing needs to plan ahead.
Mathematician Mark Lewko, commenting on Terence Tao’s blog, warned that a limit could delay public timestamping of results and leave the shared research record less complete, with more information circulating privately. That is a forecast of possible consequences, not evidence that the policy has already delayed papers.
ArXiv says it will monitor the policy and may revise it. Whether the cap reduces moderator workload, changes turnaround times or delays legitimate research remains unmeasured.
