From 1 October 2026, arXiv allows two submissions per person per calendar month, across every category. The existing cap of three active submissions at any one time, in place since 2024, stays.
This matters here because arXiv is where a large share of the work this site covers first appears — and because the reason for the limit is now the dominant fact about the research literature.
The numbers
Monthly submissions, in September:
| Submissions | |
|---|---|
| September 2016 | 9,869 |
| September 2024 | 20,569 |
| September 2026 | 40,363 |
Doubling in two years. September 2026 was the highest-volume month in arXiv’s history. It generated nearly 9,000 support tickets for staff and moderators. And in the cs.AI category alone, submissions grew more than sixfold in two years.
How the rule works
- Two submissions per person per calendar month, all categories
- Applies only to the submitting author — co-authors are unaffected
- Rejected papers still count, because they consume moderation time
- Deleting a paper before publication returns the slot
- The three-active-submissions cap is unchanged
The “rejected papers count” clause is the one that reveals what the policy is actually for. This is not a quota on publication; it is a quota on moderator attention. A rejected paper costs a moderator exactly as much as an accepted one, sometimes more, and the mechanism is designed around that cost rather than around output.
What the moderators are seeing
Thomas G. Dietterich, chair of the arXiv Editorial Advisory Council, calls the volunteer moderators the backbone of quality control, and is blunt about the distribution:
“However, 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.”
He has said on Bluesky that some authors submit dozens of papers.
The specific failure modes moderators report are worth naming, because they are each a different thing:
- Thin, narrowly focused papers
- “Salami” papers — one piece of work sliced into several submissions
- Dense texts written by AI
Dietterich also hopes the cap will curb “micro-result” papers — a submission describing one extra ablation study, or a minor architectural tweak.
That is the sharpest observation in the whole announcement, and it is not really about AI. The incentive to publish a countable unit has always existed; what has changed is that the cost of producing one fell to near zero. When writing a paper took a month, the minimum viable result was constrained by the labour. When it takes an afternoon, the floor drops to whatever is formally publishable — and a one-ablation paper is formally publishable.
A rate limit is a crude instrument against that, and it is also the only one available quickly. arXiv calls the rule a temporary fix until its own moderation tools can scale, and Dietterich says he hopes the cap can be raised or dropped later.
Three rounds of tightening in under a year
The rate limit is the third move:
- November 2025 — review articles and position papers in computer science must go through peer review before arXiv will accept them
- May 2026 — a one-year ban for authors whose papers clearly contain unverified LLM output, such as hallucinated sources
- October 2026 — two submissions per person per month
The escalation is the story. Each measure was a response to a volume that kept growing anyway.
This is not only arXiv
The conferences are in the same position, and the figures are worse.
ICLR 2027 had received about 50,000 abstracts before its deadline. ICLR 2026 got around 19,500 valid submissions. Roughly two and a half times in one cycle.
And at NeurIPS 2025, among approximately 5,000 accepted papers, GPTZero found at least 100 with fabricated citations — in papers that each had at least three reviewers. That last detail is the one to sit with: peer review, three reviewers deep, did not catch a hundred papers citing work that does not exist.
The funding, and what it is for
arXiv is splitting from Cornell University after 25 years and has been transitioning to an independent nonprofit since 1 July 2026. In late September, the Simons Foundation International, XTX Markets and the Siegel Family Endowment pledged a combined $17.2 million over three to five years.
arXiv plans to spend part of it on technical upgrades, with a focus on handling AI-generated content — which is to say the new money is going partly toward the problem the new rule is a stopgap for.
What it means if you read arXiv to find things
Practically, for anyone who uses arXiv the way this site does — scanning cs.HC and cs.GR daily for work worth covering — the effects should be mildly positive and are worth watching for:
Fewer salami submissions should mean fewer near-duplicate papers from the same group in the same week, which is one of the more time-consuming things to disentangle when deciding what is actually new.
But the cap is per submitter, not per group. A lab with six people can submit twelve papers a month without anyone exceeding the limit, and nothing in the policy changes that.
And the volume that remains is still 40,000 a month. A 2-per-author limit addresses the tail of very high-volume submitters; it does not address the median. The number of papers you cannot read is unchanged.
The more useful shift is in how to read a preprint. A paper on arXiv has been moderated, not reviewed — that was always true and is now load-bearing. The NeurIPS fabricated-citation figure says the same thing about reviewed venues. Checking that a paper’s central references exist takes two minutes and is now a reasonable default.