arXiv caps submitters at two papers a month, and rejected papers count

In September, arXiv received 40,363 paper submissions. That is double the 20,569 of the same month two years ago, and those submissions alone generated almost 9,000 support tickets. From October 1, arXiv limits each submitter to two papers a month.

arXiv is a repository where physics, mathematics and computer science papers are posted before journal review. Volunteer moderators read each submission and decide whether it meets the standards of its field. The arXiv blog says it is seeing more thin papers of narrow scope, single works split into several smaller papers, and dense AI-written papers. Submissions in cs.AI grew more than sixfold in two years.

The new rule has three parts. A submitter can send two papers a month and have at most three under review at once. A rejected paper still counts as one. A paper posted on behalf of co-authors does not use up their limit. arXiv calls the measure a stopgap until it settles on a new standard.

The part that caught my eye is that rejected papers count. It means what is being metered is not output but human review time. Making text now costs almost nothing, while reading it and sorting it still falls to people.

Open source maintainers are in the same spot. When AI-generated pull requests and bug reports pile up, the time spent filtering runs out before the time spent fixing code. Judging quality one item at a time is hard, so the answer drifts toward capping counts per account.

It is a blunt tool. A researcher with several good papers to publish, or someone working to a conference deadline, hits the same limit. That is probably why arXiv called it a stopgap. For any service whose bottleneck is the reader's time rather than the maker's, now is the time to decide where a submission limit belongs.

Sources

1. arXiv blog: Fair Moderation, Equitable Access, and AI: arXiv's Updated Rate Limit Policy, October 1, 2026.