arXiv, the open-access preprint server that underpins modern scientific communication, has imposed a hard cap of two submissions per author per calendar month, effective October 1, 2026. The policy responds to a submission volume that has doubled in just two years — from 20,569 in September 2024 to 40,363 in September 2026 — with cs.AI submissions alone surging over 6x. The moderator corps, entirely volunteer, generated nearly 9,000 support tickets from that single month's intake. The root cause is straightforward: AI authoring tools have collapsed the marginal cost of producing a paper to near zero, but the marginal cost of reviewing one remains stubbornly human. arXiv's moderators report a spike in "thin papers of narrow scope," salami-sliced research units, and dense AI-written manuscripts that meet no meaningful scholarly threshold. A small cohort of heavy submitters is consuming a disproportionate share of moderator bandwidth, delaying legitimate work by days or weeks. The two-per-month cap applies universally across all categories and all submitters. Rejected papers count against the limit — a deliberate design choice to incentivize self-curation before submission. A separate ceiling of three active submissions at any time (in place since 2024) remains. Notably, the limit applies only to the submitting author, not co-authors, creating a coordination mechanism among research teams. There are escape valves. Deleting a submission before announcement does not count against the cap. Papers held from a prior month do not consume the next month's quota (though they count toward the three-active ceiling). But the policy is explicitly described as a "stopgap" — arXiv acknowledges it needs better automated moderation tools and is buying time. The equity argument cuts both ways. For researchers in well-resourced labs with multiple co-authors who can rotate submission duties, two per month per person is rarely binding. For solo researchers, early-career scientists publishing at high cadence, or scholars in regions where a single corresponding author handles all lab output, the cap is a real constraint. The policy treats the symptom — moderator overload — without addressing the underlying incentive structure that rewards publication volume. The cs.AI category is the canary. A 6x increase in two years is not organic growth in a field; it is the signature of automated production meeting an open submission pipeline. arXiv's move implicitly concedes that disclosure-based AI policies are insufficient — authors are supposed to disclose AI use and ensure scholarly merit, but compliance is unenforceable at scale without the rate limit as a backstop. The deeper question is whether a flat per-author cap is the right mechanism or whether arXiv is deferring the harder problem of quality-based triage. Two papers per month from a productive lab is a meaningful constraint; two per month from a paper mill is still 24 papers a year per account. The policy protects moderator bandwidth in the short term but does not resolve the structural mismatch between AI-assisted production speed and human review capacity.
