arXiv has put a monthly limit on the people submitting research papers to its repository: two submissions per calendar month, across all subject categories. The policy took effect October 1, with arXiv citing a surge in submissions and the burden of low-value, AI-assisted papers on volunteer moderators. Researchers planning a batch of releases now have another constraint besides finishing the research: who submits it, and when.
The repository calls the change a stopgap while it improves moderation tools and considers practices for increasingly capable AI tools. It is not a ban on using AI in research. Nor is it a limit of two papers bearing any individual researcher’s name. That distinction matters for collaborating teams.
Two slots, not two co-authorships
The announcement’s FAQ says a submission counts only against the person submitting it, not the co-authors. The allowance spans categories; switching subjects does not provide another quota. A rejected submission still consumes a slot because the moderation work has already occurred. Deleting a submission before announcement removes it from the limits.
A separate ceiling of three active submissions, in place since 2024, remains. Papers held over from the previous month do not consume the new month’s allowance, but still occupy active slots. For example, two held papers from September would leave room for only one additional active submission, even with October’s two-slot allowance available. arXiv asks authors facing conference deadlines or a backlog to coordinate and plan accordingly.
There is an important boundary around what this process certifies. arXiv’s moderation policy explicitly distinguishes moderation from peer review. Moderators assess suitability, scholarly standards, and relevance. Appearance in the repository is not a certification that a paper’s conclusions are correct, and limiting uploads does not change that.
The scarce resource is attention
arXiv reports receiving 40,363 submissions in September, compared with 20,569 in September 2024, and almost 9,000 support tickets last month. It says moderators are seeing more thin papers, work divided into smaller publications, and dense AI-written submissions. Its diagnosis is that a relatively small group consumes disproportionate moderation time, delaying worthwhile work.
Those figures demonstrate rising workload. They do not establish that every additional paper is AI-generated or poor research. The announcement does not quantify what share of the increase falls into either category. AI can accelerate useful work as well as low-value production; arXiv still permits its use with disclosure and compliance with scholarly standards.
The strongest objection is about access, not a right to unlimited volunteer labor. In a comment on Terence Tao’s announcement post, Mark Lewko argues that a submission cap does not stop papers being written. It can instead delay the common public record while results circulate privately, making access depend more on professional connections. That is an individual argument, not evidence of a consensus among researchers.
TINA’s view
A temporary quota is a defensible response to an overloaded moderation queue. A permanent quota would need a stronger showing. Counting submissions is administratively simple, but it measures volume rather than scientific value. A third substantial result is not made less useful by arriving in the wrong calendar month. The calendar is an efficient clerk, not a scientific referee.
The submitter-only design also distributes the inconvenience unevenly. As a practical inference from the FAQ, a multi-author team can coordinate who submits its legitimate papers; a solo researcher cannot distribute that task among co-authors. That is a reason to measure the policy’s effects, not an invitation to bypass it.
Our judgment would improve if arXiv showed shorter delays for substantive work without disproportionately blocking legitimate submissions. It would worsen if the queue stayed congested while researchers increasingly withheld completed results from the shared record. The useful next signals are moderation turnaround, complaints about delayed legitimate work, and whether the promised tooling improvements let arXiv revise this stopgap. A quieter intake queue alone would not prove success.
