Too much AI slop: arXiv limits researchers to two papers a month
The moderators were overwhelmed with an influx of AI research submissions.
arXiv was operated by Cornell University before becoming an independent nonprofit in July 2026.
Follow us
- arXiv now limits each submitter to two papers per month across all categories.
- The repository says AI tools helped drive September 2026 submissions to 40,363, straining volunteer moderators.
- arXiv says low-quality submissions can delay stronger papers for days or weeks.
- The cap is temporary while arXiv updates rules and moderation tools for AI-assisted research.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
The research preprint repository arXiv has capped submissions at two per person each month across all categories, as a surge in low-quality AI spam has strained its volunteer moderators.
arXiv said that it received 40,363 submissions in September 2026, up from 20,569 submissions in September 2024, with almost 9,000 support tickets generated for its moderators.
The organization said AI tools drive much of the increase. Although AI can be used for research purposes, arXiv sets standards for the resulting work – standards that many of the new submissions don’t meet. These include scholarly standards, as well as a requirement for the work to be “of scholarly interest to the scientific community.”
According to the organization, AI has made it easy “to flood arXiv and other repositories” with “low-value papers”.
“A relatively small proportion of authors are submitting a large number of low-quality papers and consuming a disproportionate fraction of the moderators’ time. This is unfair to authors who continue to submit quality papers — their papers can be delayed for days or weeks as a result,” Thomas Dietterich, Distinguished Professor Emeritus, Oregon State University and Chair of the arXiv Editorial Advisory Council, said.
Dietterich also posted on X that some authors are submitting dozens of papers, eating up moderation resources and causing delays to papers being released.
Policy update: a battle with AI slop
The new change builds on existing rate limits, which have primarily been left to moderators’ discretion.
The repository said that the two-submission limit applies to all submissions regardless of whether they’re accepted or rejected, because each one consumes moderator time.
“We want to encourage authors to be mindful with their work, taking care that they are submitting something substantial of scholarly interest that meets arXiv’s guidelines and standards,” the organization said.
However, the limit applies per submitter, so large research groups are unlikely to face the same constraints as individual researchers. A paper with multiple authors will only count towards the allowance of the individual who submitted it.
“It is important for co-authors to coordinate their submissions accordingly,” arXiv reminded.
The current limit is a “stopgap” while arXiv determines best practices for authors using increasingly advanced AI tools and works to improve its moderation tools and procedures.
Earlier this year, the organization said it would enforce a year-long ban for authors of papers showing “incontrovertible evidence that the authors did not check the results of LLM generation,” such as hallucinated references or residual “meta-comments”.
Major computer science conferences including CVPR, ICCV, KDD, AAAI, IJCAI, and WSDM have also introduced per-author submission limits as they face an influx of AI research submissions.
NeurIPS drew 21,575 submissions in 2025, up from 9,467 in 2020, while the International Conference on Learning Representations (ICLR) received 19,525 submissions for its 2026 conference, up around 68% from 11,603 in 2025.
