Skip to content
AI

arXiv enforces two-paper monthly cap as AI spam overwhelms moderators

The preprint server limits authors to two uploads monthly as low-quality AI content overwhelms volunteer moderators and delays legitimate research.

Illustration: A massive pile of crumpled paper balls buries a small, neat stack of pristine documents.
AI-generated illustration: TechSec Headlines

Key points

  • arXiv enforces a two-submission monthly cap for all authors to manage a surge in low-quality AI-generated papers.
  • September saw record submissions, with the cs.AI category growing sixfold in two years and generating thousands of support tickets.
  • Rejected papers count toward the monthly limit, and new endorsement rules require institutional emails plus prior authorship or personal vouching.

The short version

  • arXiv now restricts each submitter to two papers per calendar month across all subject categories, effective immediately.
  • This policy adds to an existing limit of three active submissions at once, which has been in place since 2024.
  • The measure aims to reduce the burden on volunteer moderators who are overwhelmed by a flood of non-compliant AI-generated content.

The details

arXiv implemented a strict monthly submission cap this month, allowing authors to upload only two papers per calendar month. This rule applies to every subject category on the platform and operates alongside an existing restriction that limits users to three active submissions simultaneously, a policy introduced in 2024. Crucially, rejected papers count toward the monthly total because the moderation process consumes resources regardless of the final outcome. According to Socket, the repository attributes this operational strain to a massive influx of low-quality submissions generated by artificial intelligence tools.

The volume of content has surged dramatically. September recorded 40,363 submissions, a sharp increase from 20,569 in September 2024 and 9,869 a decade prior. Within the computer science discipline, the artificial intelligence category, cs.AI, experienced more than a sixfold growth in submissions over two years. This volume generated nearly 9,000 support tickets for staff and moderators last month alone. While arXiv permits the use of AI tools in research if disclosed and meeting standards, the sheer volume of non-compliant work is delaying the processing of legitimate scholarly papers by days or weeks.

The context

The escalation began in early 2025, when Ralph Wijers, then chair of arXiv’s editorial council, noted a surge in AI slop submissions first in computer science and later in other fields. He told Science that rejection rates rose from a historical average of 4% to roughly 10%–12%. Wijers described absurd examples, such as four-page papers with 50 subsections, some containing no text. Prior to this surge, moderators had expected fake papers after ChatGPT’s 2022 launch but saw no significant change through 2024.

arXiv has been tightening access since late 2024. In October, computer science moderators stopped accepting review articles and position papers without prior peer-reviewed acceptance, citing hundreds of monthly submissions that amounted to little more than annotated bibliographies. In December, the mathematics section tightened endorsement requirements, requiring more than just an institutional email. By January, arXiv extended these stricter endorsement rules to all sections, explicitly citing an unsustainable influx of non-scientific submissions. New submitters now need an institutional email plus previous authorship in the domain or a personal endorsement from an established author.

What to watch

Thomas Dietterich, chair of arXiv’s Editorial Advisory Council, highlighted the operational cost to researchers. arXiv describes the submission cap as a stopgap measure while it improves moderation processes and updates its guidance on AI use. The rules follow the individual submitter, meaning co-authors do not consume their own allowance when another person submits the paper. Wijers noted that almost all AI-generated rubbish comes from first-time submitters, which the endorsement requirement aims to mitigate by forcing new authors to get vouched for by established peers before entering the queue.

What to do and how to stay safe: arXiv

  • Monitor official arXiv announcements for updates on submission guidelines and AI disclosure requirements.
  • Ensure all AI-assisted research is clearly documented and meets the platform's scholarly standards before uploading.
  • Establish or maintain an institutional email address and secure necessary endorsements if you are a new submitter.
  • Review your submission history to ensure rejected papers do not inadvertently exhaust your monthly allowance.

Step-by-step guide: Shadow AI Mistakes That Put Corporate Data at Risk

General security guidance from the TechSec Headlines newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Does the new submission limit apply to all users or only new accounts?

The limit applies to all authors across all subject categories, restricting each submitter to two papers per calendar month.

Do rejected papers count toward the monthly submission limit?

Yes, rejected papers count toward the monthly total because the moderation process consumes resources regardless of the outcome.

What are the new endorsement requirements for new submitters?

New submitters need an institutional email address plus previous authorship in the domain or a personal endorsement from an established arXiv author.

Sources

  1. Socket
arXivAI slopsubmission limitsacademic publishingmoderation

Related stories

Shadow AI Mistakes That Put Corporate Data at Risk

Unapproved browser extensions and personal API keys route sensitive internal data to public model providers long before network firewalls notice.

Independent cybersecurity reportingGet Alerts