arXiv received 40,363 submissions in September 2026, a record month that generated almost 9,000 support tickets for its staff and volunteer moderators. On October 1, 2026, arXiv answered with a blunt instrument: an updated rate-limiting policy that caps every submitter, in every category, at two submissions per calendar month, alongside an existing limit of three active submissions at any one time. All of those volume figures are arXiv-reported, with no independent verification available yet. But the caps themselves are binding now, and if you run a lab, manage a multi-author collaboration, or oversee an institutional preprint workflow, the practical question is not whether the policy is fair. It is how to sequence your submissions so the quota accounting does not strand a paper at the wrong moment.
The short answer: treat two-per-month and three-active as hard budget lines, rotate the submitter role across co-authors, stagger backlogs across calendar months, and know that deleting a paper before it is announced is the only way to get quota back within the same month. The longer answer, including what the policy conspicuously does not do, follows.
What changed on October 1, and why
The announcement frames the policy as an explicit stopgap, not a settled position:
“As of October 1st, 2026, arXiv is instituting an updated rate limiting policy across all submitters and categories as a stopgap while we determine what may be the new best practice for authors employing more and more advanced AI tools, and to give us time to improve our moderation tools and procedures accordingly.”
Two details in that sentence matter for planning. First, “all submitters and categories” means there are no per-category carve-outs, even though arXiv’s own numbers show the pressure is uneven: according to the announcement, cs.AI submissions grew more than 6X over the past two years while other categories roughly doubled. Second, “stopgap” means the specific numbers are provisional. arXiv says it will monitor the effects and modify the policy as needed, so any workflow you build around two-per-month should tolerate the cap moving.
The growth series arXiv published is worth holding onto, because it is the entire public justification for the caps. September 2016: 9,869 submissions. September 2024: 20,569. September 2026: 40,363. Submissions doubled in two years, and the September surge produced almost 9,000 support tickets. These are arXiv’s own figures, self-reported in the policy announcement, and nothing in the available record independently confirms them. They are nonetheless the numbers arXiv will use to justify whatever comes next, so they are the numbers worth watching.
The caps in plain terms
The core mechanics, all from the announcement and its FAQ:
| Rule | What counts | Planning consequence |
|---|---|---|
| Two submissions per calendar month | Rejections included; the limit applies to submissions, not announced papers | A rejected paper still burns one of your two slots |
| Three active submissions at any time | In place since 2024; unchanged | Pipeline depth is capped regardless of monthly cadence |
| Limits bind the submitter only | Co-authors’ rates are unaffected | Collaborations can rotate who submits |
| On-hold submissions in a later month | Skip that month’s quota but still occupy the active cap | A held paper blocks one of three active slots indefinitely |
| Delete before announcement | Does not count toward either limit | The only quota refund, and only before the paper is announced |
The rejection accounting is the part most likely to surprise people. arXiv’s FAQ states it directly: “Yes – this rate limit is for submissions, not announced papers, because it is submissions that consume moderator time. If you submit two papers in a month, whether they are accepted or rejected, you will have to wait until the next month to submit to arXiv again.” The logic is internally consistent, moderator effort is spent at submission time, not at acceptance, but it raises the cost of submitting borderline work. A paper you are unsure will pass moderation is no longer a free option; it is half your monthly budget.
The three-active cap is not new. It has existed since 2024, and the October change adds the explicit monthly rate on top of it. Rate limiting itself is also established arXiv practice, previously left largely to moderator discretion. What changed is that the throttle is now uniform and numeric rather than judgment-based.
Who actually gets throttled
arXiv’s stated target is narrow: “a relatively small proportion of authors are submitting a large number of low-quality papers and consuming a disproportionate fraction of the moderators’ time. This is unfair to authors who continue to submit quality papers — their papers can be delayed for days or weeks as a result.” Moderators report increases in thin papers of narrow scope, “salami” slicing of one study into several smaller papers, and dense AI-written submissions, according to the announcement.
If that diagnosis is right, a uniform cap is a somewhat odd instrument. A researcher submitting two careful papers a month is untouched; a salami-slicer splitting one study into six papers is stopped at two. But so is a productive lab during a conference deadline cluster, or a collaboration wrapping up a grant with several finished manuscripts. The cap cannot distinguish between low-quality volume and legitimate volume, and arXiv does not claim it can. It is priced for the average submitter and binding on the tail.
There is also a complication in the causal story. arXiv’s framing ties the surge to AI-assisted writing, and the cs.AI 6X growth figure supports an association. But a bibliometric study of preprint adoption in AI policy research (arXiv:2505.03835, abstract only) attributes preprint growth between 2015 and 2024 to multiple drivers, COVID-19, ChatGPT’s release, and regional research cultures across the US, Europe, and South Korea, with significant variation in magnitude and trajectory by region. That study measures preprint citations rather than arXiv submission volume, so it cannot confirm or refute arXiv’s numbers. What it does show is that “AI flood” is at most a partial explanation for rising preprint volume. The quota treats a multi-cause phenomenon with a single lever.
The sequencing playbook
arXiv itself tells authors what to do: “We ask authors who have a backlog of work to submit, or who need to submit a paper at a specific time for a conference, to plan their submissions carefully so that the monthly submission rate limits and active submission rate limits will not negatively impact them.” What follows is our operational reading of the FAQ’s accounting rules, not an arXiv-prescribed workflow.
Rotate the submitter. The FAQ is explicit: “All arXiv rate limit policies apply only to submitters. If you submit a paper with several co-authors, that only counts towards your submission rate, not theirs.” A five-author collaboration has ten submission slots per month distributed across five people, as long as the group coordinates who submits what. This is the single most useful planning move available, and arXiv itself flags it: “It is important for co-authors to coordinate their submissions accordingly.” Decide the submitter when the paper is drafted, not when it is finished.
Stagger backlogs by calendar month. The quota resets monthly, not on a rolling window. A lab holding six finished papers cannot submit them in a burst; under the two-per-month cap with no co-author rotation, that backlog takes three months to clear. Work backward from conference deadlines: if a paper must be public by a fixed date, reserve a slot in the right month and hold it against emergencies.
Budget rejections. Because rejected submissions consume quota, front-load quality control. The calculus has shifted: an internal read-through before submission is now worth more than it was in September, because a rejection costs a slot, not just time.
Treat delete-before-announce as your only refund. “Submitters who delete their submitted paper before it is announced, whatever the reason, will not have that paper count toward their monthly submission or active submission rate limits.” If you spot a fatal error after submitting, pulling the paper before announcement preserves both quotas. After announcement, the slots are spent.
Watch the active cap separately. A paper stuck on hold does not consume next month’s two submissions, but it does occupy one of your three active slots for as long as it sits there. Three chronically held papers would freeze your pipeline entirely even with monthly quota to spare.
The question the FAQ never answers
Notably absent from the announcement is any appeal or exception process. The FAQ settles the accounting edge cases, rejections, holds, co-authorship, deletion, but says nothing about what a researcher can do if the cap blocks time-sensitive legitimate work: a doctoral thesis component with a hard deadline, say, or a rapid-response paper in a fast-moving area. Whether an exemption path exists, who adjudicates it, and how long it takes are simply not addressed in the published policy. I would not build a workflow that depends on getting an exception, because there is no documented mechanism to request one. If your situation genuinely requires more than two submissions in a month, co-author rotation is the only lever with documented support.
What the policy does not do
Two misreadings are worth correcting early.
First, this is not a new AI policy. The announcement restates the existing standard, AI tools are permitted as disclosed aids, provided the work meets arXiv’s standards of scholarly interest and advances the field, without redefining what counts as acceptable AI-assisted scholarship. The quota constrains how much you submit, not how you wrote it. arXiv explicitly says it is using the stopgap to figure out what new best practice for AI-assisted work should look like, so the substance of that question remains open.
Second, quotas are not arXiv’s only enforcement channel, and never have been. Identity and content integrity are policed separately, case by case, through withdrawal. The arXiv record shows a number-theory submission withdrawn by admins for fraudulent affiliation claims (arXiv:1407.7158), a computer-science submission withdrawn because it “contains fictitious content and was submitted under a pseudonym, which is against arXiv policy” (arXiv:1304.1836), and a general-relativity comment removed because the author name was an apparent pseudonym (gr-qc/0703020). These precedents predate the rate limits by years and operate independently of them. The practical implication: rotating the submitter role across real co-authors is legitimate and invited by the FAQ; submitting under borrowed or invented identities to evade the cap is the kind of violation arXiv has historically punished with removal, not throttling.
Will other preprint servers follow?
This is where honesty about the evidence matters. The available record contains no policy statements from bioRxiv, medRxiv, or any other preprint server responding to arXiv’s move. It is reasonable to suspect that AI-accelerated submission volume is not unique to arXiv, the drivers the bibliometric study identifies, COVID-era preprint normalization and generative AI, are not platform-specific, and a stopgap quota is a cheap, copyable instrument. But suspicion is not evidence. If you submit to multiple servers, check each one’s current policy directly rather than assuming convergence. Any claim that the moderation bottleneck is migrating across preprint infrastructure as a whole is, as of October 2026, inference.
Plan for two a month, and hold the policy loosely
The decision for labs and collaborations is concrete. Budget two submissions per submitter per calendar month, rejections included. Keep active submissions under three. Assign the submitter role deliberately at the collaboration level, stagger backlogs across months, and treat deletion before announcement as the only way to recover a spent slot. Do not count on an appeal process that has not been documented.
Hold the specifics with an open hand. Every volume figure behind this policy is self-reported by arXiv, arXiv calls the policy a stopgap it will monitor and modify, and the caps could tighten, loosen, or acquire per-category nuance. The durable part is not the number two. It is the shift underneath: volunteer moderation capacity is now a rationed resource, and monthly submission at arXiv is, for the first time, budgeted rather than an open queue. Build the sequencing habit; expect the numbers to move.
Frequently Asked Questions
Does a rejected submission count toward the monthly limit?
Yes – this rate limit is for submissions, not announced papers, because it is submissions that consume moderator time. If you submit two papers in a month, whether they are accepted or rejected, you will have to wait until the next month to submit to arXiv again.
Can co-authors submit papers to bypass the limit?
All arXiv rate limit policies apply only to submitters. If you submit a paper with several co-authors, that only counts towards your submission rate, not theirs.
How can I get my submission quota back in the same month?
Submitters who delete their submitted paper before it is announced, whatever the reason, will not have that paper count toward their monthly submission or active submission rate limits.
Join the discussion
Share a useful perspective or ask a question about this article.