Level 1 — Absolute Beginner

Scientists write papers about their work. Many of them put the papers on a website called arXiv. It is free. Anyone can read it.

Now arXiv has a new rule. One person can send only two papers each month. One person can have only three papers waiting at one time.

Why the new rule? In September 2026, arXiv got 40,363 papers in one month. That is a record. Two years before, in September 2024, it got 20,569 papers. So the number is now about two times bigger.

People use AI tools to write papers fast. Some of these papers are not very good. A small group of people sends many papers. The workers at arXiv do not have time to check them all.

paper

a piece of writing in which a scientist explains new research

website

a place on the internet where you can read or see things

free

costing no money

rule

something you must or must not do

record

the highest number ever reached

tool

something you use to do a job

check

to look at something to see if it is correct

limit

the largest amount that is allowed

Level 2 — Elementary

arXiv is the website where physicists, mathematicians and computer scientists post their research before it is formally reviewed by a journal. Papers posted there are called preprints, and they are free for anyone to read. On October 1, 2026, arXiv introduced a strict new limit on how much one person can post.

Each submitter may now upload a maximum of two papers per calendar month, and may have no more than three submissions active at any one time. Previously there was no monthly cap at all.

The numbers behind the decision are striking. In September 2026 arXiv received 40,363 submissions in a single month, the highest figure in its history. In September 2024 it received 20,569, and in September 2016 only 9,869. The computer science and artificial intelligence section alone grew six times larger in two years.

arXiv says the growth is driven partly by the ease of producing papers with AI tools. Moderators report more thin papers of narrow scope, and more salami papers, where one piece of work is sliced into several separate submissions. Thomas Dietterich, chair of the Editorial Advisory Council, said a relatively small proportion of authors submit a large number of low quality papers and consume a disproportionate share of the moderators' time. Authors facing conference deadlines are simply advised to plan their submissions carefully.

preprint

a research paper shared publicly before a journal has formally reviewed it

submitter

the person who sends something in for publication or consideration

cap

an upper limit on an amount or number

striking

very noticeable or surprising

moderator

a person who checks submitted material before it is published

narrow scope

covering only a very small part of a subject

proportion

a part of a whole, measured against the total

disproportionate

much larger than is reasonable or fair

Level 3 — Intermediate

For more than three decades arXiv has operated on an unusually generous premise: that any researcher who wants to share a paper should be able to, immediately and without charge, before any journal has passed judgment on it. On October 1, 2026, that openness acquired a boundary. Every submitter is now restricted to two papers per calendar month and to three active submissions at any one time.

The justification is arithmetic. September 2026 brought 40,363 submissions, a monthly record, against 20,569 in September 2024 and 9,869 in September 2016. Doubling in two years is not organic growth in the research community, which expands far more slowly than that, and the computer science and artificial intelligence category alone swelled sixfold over the same period.

arXiv attributes much of the surge to the falling cost of producing a plausible looking manuscript. Large language models can draft a paper in an afternoon, which changes the economics of submission: when writing is cheap, the rational move for an author chasing citations is to submit more. Moderators describe two characteristic symptoms. The first is the thin paper of narrow scope, technically competent but contributing almost nothing. The second is the salami paper, in which a single study is sliced into multiple submissions to inflate a publication list.

Thomas Dietterich, who chairs arXiv's Editorial Advisory Council, located the problem in a minority: a relatively small proportion of authors are submitting a large number of low quality papers and consuming a disproportionate fraction of the moderators' time. That diagnosis explains the shape of the remedy. A flat per person cap targets volume rather than content, and it does so without requiring moderators to adjudicate quality case by case. The cost falls on the unusually productive legitimate author, and arXiv's guidance to anyone caught by a conference deadline is notably unsympathetic: plan your submissions carefully.

premise

an idea or principle that something is based on

pass judgment

to form and state an opinion about the quality of something

organic growth

natural, gradual increase rather than growth from an outside cause

swell

to grow larger in size or number

plausible

seeming reasonable or believable

citation

a mention of one piece of research inside another, used to measure influence

inflate

to make something seem bigger or more important than it is

remedy

a way of solving a problem

Level 4 — Advanced

arXiv's decision, effective October 1, 2026, to cap each submitter at two papers per calendar month and three concurrently active submissions marks the first substantive retreat from the repository's founding bargain. That bargain held that the marginal cost of hosting a paper was negligible and the marginal benefit of open circulation large, and it therefore declined to ration access. Thirty years on, the first half of the proposition has quietly failed, not because storage became expensive but because human attention did.

The quantitative case is difficult to dispute. September 2026 produced 40,363 submissions, an all time monthly high, measured against 20,569 in September 2024 and 9,869 in September 2016. A near doubling across twenty four months bears no relation to the growth rate of the global research workforce, and the concentration is diagnostic: the computer science and artificial intelligence classification expanded roughly sixfold over the same interval. Where submissions rise faster than researchers, the variable that has changed is the cost of production.

Generative models have collapsed that cost. Drafting a syntactically polished, superficially novel manuscript is now a task of hours rather than months, which inverts the incentive structure that once disciplined academic publishing. When composition was laborious, scarcity of effort did the filtering; when it is trivial, nothing does. Moderators have identified the two resulting pathologies with some precision. One is the thin contribution, methodologically unobjectionable and intellectually inert. The other is salami slicing, the deliberate fragmentation of a unitary result into several nominally independent submissions, a practice that converts one finding into several lines on a curriculum vitae.

Thomas Dietterich, chair of the Editorial Advisory Council, framed the problem distributionally rather than universally, observing that a relatively small proportion of authors are submitting a large number of low quality papers and consuming a disproportionate fraction of the moderators' time. The remedy follows logically from that framing, and its elegance is also its weakness. A uniform numerical ceiling throttles the heaviest submitters without obliging moderators to arbitrate merit, an adjudication that would be both slow and contestable. But it is indifferent to quality by design, and so it constrains the prolific empiricist exactly as firmly as the opportunist. arXiv's counsel to those inconvenienced, to plan their submissions carefully, concedes the point without softening it.

substantive

important and dealing with the real substance of a matter, not a formality

marginal cost

the additional cost of producing or handling one more unit

ration

to limit the amount of something that each person may have

diagnostic

revealing the nature or cause of a problem

invert