arXiv Limits to Two Monthly Submissions Amid AI Surge

arXiv will restrict submissions to two per person per month while maintaining its cap of three active submissions. This measure follows a historic peak in deposits and a rise in AI-generated texts, as the platform becomes independent and receives $17.2 million in pledges.

Warning Signals on Work Quality

The sector is facing unprecedented volumes and quality discrepancies. At NeurIPS 2025, out of approximately 5,000 accepted papers, GPTZero identified at least 100 fabricated citations, even though each publication had been reviewed by at least three examiners. The pressure upstream is also evident in conferences: before the deadline, ICLR 2027 received nearly 50,000 abstracts, compared to about 19,500 valid submissions for ICLR 2026. A test conducted in collaboration with Princeton and the UK AI Safety Institute involved having Claude Opus 4.8 work for six days on questions derived from two unpublished NeurIPS papers; the original authors, acting as reviewers, rejected both sets of responses.

Two Submissions Per Month and Specific Rules for Authors

Starting October 1, 2026, arXiv will only allow two submissions per person per calendar month. The existing cap of three active submissions in parallel, in place since 2024, remains unchanged. This restriction applies to all categories and targets only the submitter, without impacting co-authors. Rejected papers will still count as they require moderation work, but the withdrawal of a submission by its authors before publication will free up the corresponding slot.

Record Volume and Growing Moderation Burden

In September 2026, arXiv recorded 40,363 submissions, its highest monthly total. Between September 2016 and September 2024, the number of submissions rose from 9,869 to 20,569, then surpassed 40,000 per month, doubling in two years, with more than a sixfold increase in volume in the AI category alone during the same period. This evolution has led to the creation of nearly 9,000 support tickets for staff and moderators. According to arXiv, this increase is partly due to the use of AI tools, while volunteer moderators are noticing a rise in poorly developed articles, fragmented publications, and dense texts produced by AI.

Successive Tightenings Before the New Cap

In November 2025, the platform strengthened its rules for computer science by subjecting review and position papers to peer review. In May 2026, Thomas G. Dietterich announced a one-year ban on authors whose texts contain unverified language model results, such as hallucinated sources. The chair of the Editorial Advisory Board states that volunteer moderators are the backbone of quality control and emphasizes that a minority of authors are submitting low-quality papers, taking up a disproportionate amount of moderators' time and delaying quality work by several days or weeks. He also notes that some authors submit dozens of papers.

arXiv Becomes Independent and Receives $17.2M for Modernization

After 25 years at Cornell University, arXiv transitioned to an independent nonprofit structure on July 1, 2026. By the end of September, the Simons Foundation International, XTX Markets, and the Siegel Family Endowment pledged a total of $17.2 million over three to five years. The platform plans to allocate part of these funds to technical upgrades, with a focus on managing AI-generated content. In August, Google DeepMind expanded Co-Scientist to plan experiments, operate instruments, and generate manuscripts; Anthropic is building a biology lab where Claude will guide robots; and in September, OpenAI announced an automated research intern operated under human supervision.