arXiv, the preprint server on which much of physics, mathematics and computer science is first posted, has begun rationing. From 1 October, each submitter may file at most two submissions per calendar month.
The change was announced on 1 October by Kat Boboris, arXiv's community engagement manager, who wrote that "arXiv now limits submitters to up to two submissions per calendar month". A separate cap of three active submissions at any one time, in place since 2024, remains.
The numbers behind the decision are not subtle. arXiv received 40,363 submissions in September 2026. In September 2024 it received 20,569, and in September 2016 it received 9,869. That is a doubling in two years and an increase of 309 per cent over the decade.
The server attributes the surge to low-quality AI-assisted work: papers that are thin, narrow in scope, or single pieces of research fragmented into several submissions. Moderation is done by volunteers.
"A relatively small proportion of authors are submitting a large number of low-quality papers and consuming a disproportionate fraction of the moderators' time," said Thomas Dietterich, chair of arXiv's editorial advisory council.
The limit applies regardless of whether a paper was written by a human alone or with machine assistance. A spokesperson said arXiv would ask researchers "to prioritize their submissions and consider submitting papers that reflect the full scope of their work rather than multiple incremental papers".
Papers that are rejected count towards the monthly total. On a paper with several authors, the submission counts only against whoever files it, and arXiv expects large institutions to see little practical effect.
arXiv became an independent nonprofit over the summer, after twenty-five years hosted at Cornell University. It describes the two-a-month rule as a stopgap while new best practices are worked out, and says it will monitor the effect and adjust as needed.
The rule took effect immediately, rather than at the start of a notice period, which means the allowance now governing October was announced on the first day of October.
It is worth being precise about what the rule counts, because it does not count papers. It counts submissions, and a rejected submission counts in full.
The quota is therefore not a limit on how much science a researcher may contribute. It is a limit on how many times a researcher may try. The author whose work is accepted twice a month and the author whose work is rejected twice a month draw down the identical ration.
The arithmetic follows. Two attempts a month is twenty-four attempts a year, and that allowance is the same for everybody who has ever used the service. Nor does it roll over: a calendar month is a calendar month, and a third idea arriving in October must wait until November, where it will compete with whatever November produces.
The Department of Unnecessary Research has worked through the consequences for the scientific record. Science is now a metered service on a monthly standing allowance, and the quantity of knowledge a field may generate in a year is its headcount multiplied by twenty-four.
Growth remains available, but in one direction only. A discipline that wishes to publish more must recruit. A discipline that recruits nobody acquired a fixed ceiling on 1 October.
Moderator time, which was the thing actually in short supply, has not been saved. It has been rescheduled. The papers are not fewer. They are later. January arrives with a full allowance.