The world’s largest scientific preprint server, arXiv, has limited the number of submissions a single author can send to the platform starting in October 2026. The reason is a sharp increase in the flow of submissions, which the administration says is partly linked to the spread of advanced AI tools. Moderators have faced a rise in low-quality, overly narrow, and AI-written papers, whose processing is taking more and more time.

From now on, one author may submit no more than two papers to arXiv per calendar month. At the same time, they may have no more than three active submissions under review.

The restrictions apply to all authors and all subject categories.

The Number of Submissions Has Doubled in Two Years

The scale of the problem is clearly visible in arXiv’s statistics. In September 2016, the platform received 9,869 submissions. By September 2024, that number had grown to 20,569, and by September 2026 it had already reached 40,363.

Over two years, the flow of submissions has nearly doubled. In September 2026 alone, as a result, staff and volunteer moderators received almost 9,000 support requests.

For arXiv, the increase in the number of submissions has become a problem not only because of the sheer volume of texts. Each submission requires time to review, and moderators’ capacity is limited.

That is why the new system limits the number of submitted papers rather than published ones. If a paper is rejected, it still counts toward the monthly limit: it has already required moderators’ time.

AI Has Accelerated Paper Preparation — and Their Flow

arXiv acknowledges that modern AI tools can speed up certain stages of scientific work. At the same time, however, they have lowered the barrier to the mass creation and submission of scientific texts.

Moderators note a rise in several types of problematic materials at once. Among them are “thin” papers devoted to extremely narrow questions, so-called “salami papers,” where the results of a single study are artificially split into multiple publications, as well as texts created using AI.

The problem is especially acute among a small group of authors. According to arXiv’s estimates, a relatively small number of people submit a large volume of low-quality work and thus consume a disproportionately large share of moderators’ time.

“This is unfair to researchers who continue to submit high-quality work,” the platform said in a statement.

At the same time, arXiv is not introducing a general ban on the use of AI in preparing scientific publications. The problem for the platform is not the mere fact of using such tools, but the sharp growth in the number of submissions, among which there has been a greater share of materials that do not meet its standards.

How the New Limit Works

In addition to the cap of two submissions per month, there is also a limit of three active submissions at the same time.

If a paper was submitted in the previous month and is still under review, it does not reduce the new monthly limit in the following month. However, it still counts toward the cap of three active submissions.

There is also an exception for withdrawn materials. If an author withdraws a paper before it has been announced, it does not count toward either the monthly limit or the overall cap on active submissions.

In this way, arXiv is trying not simply to reduce the overall flow of submissions, but to redistribute moderators’ limited time so that it is not spent primarily on a small group of the most active authors.

What Will Change for Scientists

arXiv remains one of the most important tools for the rapid dissemination of scientific results. Unlike traditional journals, preprints allow researchers to publish their work before completing the lengthy peer-review process.

The platform has existed since 1991 and today contains more than 3 million papers in physics, mathematics, computer science, biology, statistics, economics, and other fields.

That is why the new limit may become a noticeable change for researchers who regularly publish a large number of results. However, judging by arXiv’s explanation, the main goal of the policy is not to restrict scientific activity as such, but to cope with a new reality in which producing text has become significantly cheaper and faster.

For scientific infrastructure, this creates a paradox: AI can accelerate the research process itself and the preparation of publications, but at the same time it can multiply the number of materials that humans have to review. As a result, automating one part of the scientific process creates an additional burden on another — the one where human expertise remains necessary for now.