arXiv has imposed strict limits on submissions because volunteer moderators can no longer handle the volume of AI-generated content. Researchers can now submit a maximum of two papers per calendar month and may have no more than three active submissions at any time. This follows a recent blog post stating that artificial intelligence has made it too simple to flood the publisher’s inbox.
The bottleneck
Thomas Dietterich, a professor emeritus at Oregon State University and chair of arXiv’s editorial advisory council, explained that the system relies on unpaid volunteers to filter low-quality work. He noted that a small group of authors is consuming a disproportionate amount of the moderators’ time by submitting large numbers of poor papers. This situation delays quality research for days or weeks.
The numbers show a steep climb in activity. The repository received 9,869 submissions in September 2016. That figure rose to 20,569 in September 2024, and reached 40,363 in September 2026. Submissions across the site have doubled in the last two years, while the computer science category has seen a six-fold increase.
How the surge happens
Researchers are using large language models to parse data, write code, and design experiments. Some are also using these tools to draft the papers themselves. arXiv states that its policy allows the use of generative AI as long as authors disclose it and the work meets scholarly standards.
The archive reports a rise in “thin papers of narrow scope” and “salami papers,” where researchers split one study into multiple articles. There is also a marked increase in dense, AI-written documents. The publisher says these tools make it easy to flood the site with low-value content.
Previous restrictions
Rate limiting existed before, but moderators previously applied it at their own discretion. The new rules aim to distribute the workload of these volunteers more fairly. This update is part of a longer conflict with AI-generated slop and spam.
Less than a year ago, the archive stopped accepting review articles and position papers in computer science, citing low-effort LLM writing as the cause. In January, it began requiring new authors to have an endorsement from an existing user. Previously, an academic email address was the only requirement. In May, the site announced a one-year ban for researchers who submitted AI slop.
What it means
The change forces researchers to curate their output rather than publish everything immediately. Authors must wait for previous submissions to be processed or rejected before sending new work. This slows down the influx of papers but aims to protect the integrity of the repository for genuine contributors.
