arXiv, the open-access repository where scientists share research before formal publication, began limiting submitters to two papers per calendar month on Thursday, a stopgap meant to slow a surge of thin, A.I.-assisted submissions that it said was consuming its volunteer moderators’ time.
The numbers explain the move. arXiv received 9,869 submissions in September 2016 and 20,569 in September 2024. This September the figure reached 40,363, double the level of two years earlier, and the month generated nearly 9,000 support tickets for arXiv staff and moderators.
The new rules apply across every subject category, and rejected papers count toward the monthly total, because it is submissions, not announced papers, that consume moderator time. A separate cap of three total active submissions, in place since 2024, stays. Papers still on hold when a new month begins do not count against that month’s quota but do count toward the active limit, and papers deleted before they are announced count toward neither. The limits bind whoever submits, not co-authors, and the site asked authors facing conference deadlines or backlogs to plan accordingly.
Behind the change is a shift in what arrives. arXiv policy permits A.I. as a tool in support of research, as long as its use is disclosed and the work meets the site’s standards of scholarly interest. Lately, though, moderators report a rise in thin papers of narrow scope, in “salami” papers that slice one piece of work into several smaller submissions, and in dense A.I.-written manuscripts. “A relatively small proportion of authors are submitting a large number of low-quality papers and consuming a disproportionate fraction of the moderators’ time,” Thomas Dietterich, a distinguished professor emeritus at Oregon State University and chair of the arXiv Editorial Advisory Council, wrote in the announcement. “This is unfair to authors who continue to submit quality papers — their papers can be delayed for days or weeks as a result.”
Before A.I., arXiv said, the human pace of research imposed a practical ceiling on submissions, which moderators enforced at their discretion. That ceiling is gone, and the cap buys time to work out best practices for authors using ever more advanced tools and to improve moderation tools and procedures to match. The site said it would monitor the effects on authors, readers and the corpus, and modify the policy as needed.
Reactions from working scientists pointed at the strain from both directions. “ArXiv has become too slow to publish,” one commenter, vzaliva, wrote on Hacker News, saying one paper had been “on hold for 18 days and counting” while another, posted to the repository hal.science instead, “was up in a day.” Another commenter, totetsu, put the problem more broadly: “Something that was a common good, that worked at a human scale, is now made unviable because automation has pushed it beyond what it can handle and still be useful.”