arXiv's Strictest Rate Limit: Two Papers a Month as the AI Submission Flood Peaks

The New Rule: Two Papers a Month, No Refunds on Rejection

From October 1, preprint platform arXiv enforces an updated rate-limit policy across all submitters and categories: at most two submissions per person per calendar month, and at most three active submissions at any time. The sharpest detail: quota is counted per submission — a rejected paper still burns the slot. The two-submission quota is shared across all of arXiv (switching categories does not help), and only submissions withdrawn by the author before official announcement are refunded.

arXiv's official blog is explicit that this is a stopgap measure while it upgrades moderation tools and procedures.

The Paper Flood: Doubled in Two Years, AI Track Up More Than 6x

The immediate driver is the surge in submissions. In September 2026, arXiv received 40,363 submissions in a single month — an all-time record — generating nearly 9,000 support tickets along the way. The official post names three kinds of problem submissions: thin papers of narrow scope, "salami" papers that slice one piece of work into many, and a marked increase in dense AI-written papers.

arXiv's stance is not conservative: AI assistance is permitted with disclosure, but AI has drastically lowered the barrier to producing low-value papers. In the blog's own words — "before the advent of AI, there was an easily discernible, practical limit" to how many submissions one person could produce.

Papers Can Scale; Moderators Cannot

Thomas Dietterich, chair of the arXiv Editorial Advisory Council and distinguished professor emeritus at Oregon State University, notes that a small share of authors submitting large volumes of low-quality papers consumes a disproportionate fraction of volunteer moderators' time — delaying quality submissions by days or even weeks.

This is not an isolated case. In February, Nature reported how HCI researcher Raphael Wimmer used AI tools to write up an experiment he never actually ran. When paper production starts to scale while gatekeeping headcount does not, rate-limiting at the entrance becomes almost the only option.

The practical impact on researchers is just as concrete: there is no fast lane for conference deadlines or backlogs — the official advice is to plan ahead and "submit only your best work." A multi-author paper consumes only the submitter's quota, so co-authors must coordinate who clicks the button; and since moderation still relies on volunteers, rate limiting is essentially about distributing scarce review time more fairly.

The deeper impact sits at the ecosystem level: as preprint servers become the primary outlet for LLM-driven "paper farms", arXiv's citation and distribution signals are being diluted, and downstream conferences, journals and literature surveys all face more noise. Building new quality signals for research with heavy AI participation may be the question the whole publishing industry must answer after the rate limit.

Takeaway: AI Content Governance Is a Shared Exam for Every Industry

arXiv's predicament is a dress rehearsal for the whole content ecosystem: as generation cost approaches zero, quality control must move upstream. Enterprises need the same foresight in AI-generated content labeling, audit and compliance — the exact "easy to generate, hard to verify" scenarios addressed by NineZenith's Zenith-Safety (Tiandun) product line.

(Compiled from the arXiv official blog, QbitAI, Nature and other public reports)