arXiv will limit submissions to two per person per month starting October 2026. The policy responds to a surge in submissions, especially in AI, where monthly submissions have grown more than sixfold in two years. arXiv says a small fraction of authors are flooding the platform with low-quality papers, overwhelming volunteer moderators.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Starting October 1, 2026, arXiv caps submissions at two per person per month, while keeping the existing rule from 2024 limiting authors to three active submissions at once. This is not routine governance, but a direct response to volunteer reviewers' capacity after submissions doubled in two years and cs.AI grew more than sixfold in two years.
Starting October 1, 2026, arXiv sets a cap of two submissions per person per calendar month. This cap applies to all categories and targets only the submitter themselves; co-authors are unaffected. The older cap in place since 2024 limiting authors to three active submissions at once continues, and the two rules run in parallel.
The rule's treatment of "rejected papers" is worth noting: rejected papers still count toward the cap because they consume review time; but if authors delete a paper before publication, they can reclaim that slot. This means the cap constrains the act of submission itself, not the final publication outcome.
arXiv's monthly submissions rose from 9,869 in September 2016 to 20,569 in September 2024, then to 40,363 in September 2026. The cs.AI category alone grew more than sixfold in two years. September 2026 was the highest single-month submission volume in arXiv's history.
This surge generated nearly 9,000 support tickets for staff and reviewers. Volunteer reviewers reported seeing more thin, narrowly scoped papers, "salami-sliced" papers cut into small pieces, and dense text written by AI.
Thomas G. Dietterich, chair of arXiv's Editorial Advisory Committee, called volunteer reviewers the backbone of quality control and noted that a small number of authors submitting large volumes of low-quality papers consumes too much reviewer time, which is unfair to authors who consistently submit high-quality papers, whose papers may be delayed by days to weeks as a result. Dietterich added on Bluesky that some authors submitted dozens of papers.
Dietterich also hopes the cap will curb "micro-result" papers, such as those describing only one additional ablation experiment or a minor architectural tweak. This framing extends the cap's goal from "reducing quantity" to "changing incentives."
arXiv calls the rule a temporary remedy until its own moderation tools can scale, and Dietterich said he hopes the cap can later be raised or removed. But the source gives no timeline for when the moderation tools will scale or when the cap will be adjusted or removed.
Whether the cap can truly curb low-quality papers and "micro-result" papers remains unclear; the source also provides no actual data on the policy's improvement of submission delays for high-quality authors. Therefore, the conclusion that the two-per-month cap has solved the problem currently lacks evidentiary support.
arXiv had previously tightened its computer science paper rules in November 2025, requiring survey articles and position papers to undergo peer review; in May 2026, Dietterich announced a one-year submission ban for authors whose papers clearly contain unverified LLM output (such as fabricated citations). Together with this two-per-month cap, this is the third round of tightening in under a year.
This sequence shows arXiv's governance focus shifting from "post-hoc punishment" to "ex-ante throttling," but the two-per-month cap itself is still officially positioned as a temporary measure.
AI conferences face the same problem: ICLR 2027 had received about 50,000 abstracts before the deadline, ICLR 2026 received about 19,500 valid submissions; among the roughly 5,000 accepted papers at NeurIPS 2025, GPTZero found at least 100 fabricated citations, despite each paper having at least three reviewers.
The causal link between these data and arXiv's submission surge is not directly argued by the source, so they can only be presented as contemporaneous phenomena in parallel, not inferred as direct results of the same mechanism.
After 25 years, arXiv separated from Cornell University and became an independent nonprofit starting July 1, 2026; in late September, Simons Foundation International, XTX Markets, and Siegel Family Endowment pledged a total of $17.2 million over three to five years, and arXiv plans to use part of the funds for technical upgrades, with a focus on handling AI-generated content.
Meanwhile, major AI labs are aggressively positioning their models as research tools: in August, Google DeepMind expanded its Co-Scientist into a system that can plan experiments, control lab equipment, and generate manuscripts; Anthropic is building its own biology lab to let Claude guide robots in experiments; in September, OpenAI announced an "automated research intern" that handles established research tasks under human supervision. The actual impact of these tools on submission volume has not been quantified, and the specific amount of the $17.2 million devoted to handling AI-generated content and its effects are also unstated.
The facts in this article mainly come from THE DECODER's report on arXiv's policy announcement, which is secondhand retelling. The two-per-month cap, retention of the old cap, scope of application, rejected-paper counting rule, submission volume figures, cs.AI growth, support ticket count, Dietterich's statements, and conference-side data are all presented as stated by the source. arXiv calling the rule a "temporary measure" and Dietterich hoping to later raise or remove the cap are official statements quoted by the source; this article labels them as source_claim and does not treat them as fulfilled promises. Whether the cap can truly curb low-quality papers, when the moderation tools will scale, and the specific amount of the funding devoted to AI content handling and its effects are all not given by the source, and this article preserves them as uncertainties.
The two-per-month cap is a temporary gate arXiv chose after submissions doubled in two years, cs.AI grew more than sixfold, and volunteer reviewers were overwhelmed by nearly 9,000 support tickets. It constrains submission behavior rather than publication outcomes; rejected papers still occupy a slot, and only deletion returns it. The real variables are when arXiv's moderation tools will scale, whether the $17.2 million in funding can translate into capacity to handle AI-generated content, and where submission volume will go after AI labs push their models as research tools—the source gives no answers to these.
Primary report
Primary source