The New Policy
Effective 1 October 2026:
- Two submissions per calendar month, per submitter
- Three active submissions at any one time
- Applies to all submitters in all categories, with no subject-area exceptions
- Papers deleted before announcement do not count, so revising costs nothing
The Numbers Behind It
| Metric | Figure |
|---|
| Submissions, September 2026 | 40,363 |
| Submissions, September 2024 | 20,569 |
| cs.AI category growth, two years | 6x |
| Support tickets from September alone | nearly 9,000 |
Submissions roughly doubled in two years across the whole archive. The AI category grew six times over. arXiv is run with volunteer moderators, and nine thousand support tickets in a month is the number that actually forced this.
What arXiv Says Is Wrong With the Submissions
Three specific problems are named: thin papers, salami papers — one piece of work sliced into several submissions to inflate a publication count — and dense AI-written papers that do not meet the bar for scholarly interest.
Notably, arXiv has not banned AI assistance. Authors may use AI tools provided use is disclosed and the work meets arXiv's standards. The objection is to output that fails the standard, not to the tool.
That distinction is doing real work, and it is the right one. A rule against AI-assisted writing would be unenforceable and would catch honest researchers. A rule against submissions that are not interesting enough to be worth a moderator's time is a quality standard that happens to catch most of what the tooling produces.
Two Papers a Month Is a Blunt Instrument, and arXiv Knows It
A flat cap punishes the prolific alongside the abusive. A large lab publishing genuinely across several subfields, or a senior researcher who is a named author on many submissions, now has a ceiling that has nothing to do with the quality of their work.
arXiv is fairly direct that this is a holding action — the policy exists to buy time to improve moderation tools and procedures. Read as a permanent rule it is crude. Read as a rate limiter while better filtering gets built, it is a reasonable call made by people with no good options.
The Pattern This Belongs To
This is the same shape as several stories this fortnight. Volume that was previously constrained by human effort is no longer constrained, and institutions built around the old constraint are discovering their defences were economic rather than technical.
Peer review, moderation queues, submission processes, inbox filters, code review: all of them assumed producing a plausible artefact was expensive. When it stops being expensive, the queue is the first thing to break — and the first fix is always a rate limit, because it is the only one you can ship in a week.
What It Means Practically
- If you publish to arXiv: plan the calendar. Two a month, three active, and deletions before announcement are free.
- If you read arXiv: expect volume to drop and the average to rise, at least until people adapt.
- If you build on arXiv data: your ingestion volume changes from October. Anything calibrated on 2026 submission rates needs rechecking.
FAQ
What is the new arXiv limit?
Two submissions per calendar month and three active submissions at a time, for all submitters in all categories, from 1 October 2026.
Does it ban AI-written papers?
No. AI tools may be used if disclosed and the work meets arXiv's standards.
Why now?
September 2026 brought 40,363 submissions against 20,569 in September 2024, with cs.AI up sixfold in two years and nearly 9,000 support tickets in one month.
Do withdrawn papers count?
Papers deleted before announcement do not count toward the limit.