arXiv announced a limit of two new submissions per month: what has changed for authors

arXiv publishes scientific preprints without full peer review. The archive’s announcement states that starting in 1 October 2026, each submitter will be able to submit no more than two new papers per calendar month, and that the limit is to apply regardless of scientific category or moderation outcome. At the same time, an author may have no more than three active submissions.

The announcement calls this decision a temporary measure. The archive links it to the workload of volunteer moderators and the development of tools and procedures for reviewing publications.

What limit applies to arXiv authors

According to the rule’s description, it applies specifically to new submissions, not only to texts published in the archive’s listings. If an author submits two papers in a month, a third submission will become available after the start of the next calendar month. arXiv materials also state that rejecting a paper does not return the attempt used.

The limit is counted at the submitter level. When one researcher uploads a paper on behalf of several coauthors, the submission counts toward that researcher’s monthly limit but does not reduce the coauthors’ limits. This arrangement requires duties to be allocated within the research group in advance, especially when several materials are nearly ready for publication at the same time.

• New submissions are limited to two per calendar month.
• The total number of active submissions must not exceed three.
• According to the rule’s description, it applies to all arXiv categories.
• A rejected submission counts toward the monthly limit.
• A paper deleted before the announcement is not counted toward the monthly or active quota.

Separate rules apply to replacements of already published versions. For replacements after the fifth version, the limit is one replacement per week. Authors need to distinguish between a new publication, an active submission, and an update to an existing paper: the arXiv description states that these actions do not always use the same limit.

Why the publication volume has grown

According to data cited by arXiv, the archive received 20 569 submissions in September 2024, while in September 2026 the number reached 40 363. Over two years, the monthly volume approximately doubled. At the same time, staff and moderators received nearly 9 000 inquiries related to processing submissions and operating the archive.

The growth is unevenly distributed across categories. In artificial intelligence, the number of submissions during the specified period increased more than sixfold, whereas in other categories the growth, according to the archive’s data, was closer to twofold. These figures describe the volume of incoming materials, but by themselves do not prove that the quality of all scientific output has declined.

arXiv links part of the acceleration to more accessible generative artificial intelligence tools, which simplify certain aspects of research work. The archive’s announcement also says that such tools may contribute to the emergence of a large number of materials with limited scientific value. This explanation concerns one factor behind the workload and does not describe the quality of all incoming papers.

How generative AI has made moderation more difficult

According to arXiv’s reference materials, moderation takes place before a paper is publicly announced. It is performed by volunteer specialists who check whether the material fits the scientific subject area, meets formatting requirements, and satisfies basic standards of academic communication. Moderation does not replace peer review: the archive’s decision does not mean that the publication has undergone an independent expert review of all its methods, data, and conclusions.

The archive reports several types of problematic materials. These include narrow texts with insufficient substantive contribution, publications created by artificially splitting a single study, and stylistically dense papers prepared with substantial assistance from generative tools. The moderation description also mentions checking originality, novelty, significance, and the presence of substantive research.

Why has this become a bottleneck? A generative system can quickly create coherent text, but coherence does not confirm novelty, experimental correctness, or the reliability of references. If the flow of such materials grows faster than the number of qualified moderators, conscientious authors also face delays because their papers pass through the same queue. This is an inference from the workload described by arXiv, not a separate measurement of delays for each category of author.

Splitting research and artificially expanding it

ArXiv mentions publications created by artificially splitting a single study among the problematic materials. This description points to one source of the workload but does not establish a universal criterion under which every series of papers within one research program should be considered a violation.

Several texts may concern the same area, and evaluating their content depends on the specific material. The arXiv article discusses publications in which splitting the research results in insufficient substantive contribution or excessive repetition, but the detailed procedure for such an assessment is not disclosed in the materials provided.

The archive separately mentions narrow texts and papers written with substantial assistance from generative tools. These categories may overlap, but the materials provided do not describe a specific technique for artificially inflating the volume of a small result.

Can AI be used when preparing a paper

arXiv’s reference materials state that generative language tools may be used as aids. Authors must disclose substantial use of such tools in the paper, taking into account the standards of their scientific field. Artificial intelligence cannot be listed as an author because it does not assume responsibility for the publication’s content.

Permission to use AI does not mean permission to automatically submit unchecked text. Authors need to independently verify facts, mathematical calculations, code, tables, references, the description of the experiment, and consistency with the stated data. Particular attention is required for errors that appear convincing and can therefore pass ordinary editorial proofreading.

1. Record which tools were used and for what tasks.
2. Check every reference and claim against the original scientific materials.
3. Compare the results in the text with the original data and code.
4. Make sure the paper contains an independent and clearly explained scientific contribution.
5. Disclose substantial use of generative AI in accordance with the requirements of your field.

Who is responsible for errors and data

Responsibility remains with the people who sign the paper. If a language model invented a reference, distorted a result, introduced plagiarism, or formulated a false conclusion, the authors are responsible for including that passage in the publication. Responsibility is not shifted to the model’s developer and does not disappear because the text was generated automatically.

This rule also applies to coauthors. Before submission, they must agree on the fact of submission, the authorship, and the affiliations. The submitter must have permission to post the material, and the listed authors must be represented correctly.

Why the limit is considered temporary

The limit reduces the number of submissions reaching moderators over a short period, but does not in itself determine the quality of scientific papers. Two strong papers and two weak submissions use the same number of attempts. The archive describes the quota as an interim solution intended to allow time to improve moderation tools and revise procedures.

A long-term system will need to distinguish between technical checks, subject classification, duplicate detection, and assessment of substantive contribution. Automated tools may help find similar texts, check metadata, and refer complex cases to specialists, but final scientific judgment cannot be reduced entirely to a machine filter. An algorithmic error is especially consequential when a rejected submission also uses up the author’s limited opportunity for the next publication.

Authors with a portfolio of completed papers should plan submissions in advance, take active submissions into account, and not upload a text until the review is complete. Research groups should determine in advance who will submit the material and when a new version will be needed. For conference deadlines, this leads to a practical recommendation: allow extra time, since an urgent submission may run into the monthly quota.

What changes for researchers

According to arXiv’s description, the new rule does not prohibit publishing results and does not ban the use of generative AI. It changes the pace of submissions and raises the cost of an ill-considered submission. Authors must choose materials that genuinely meet arXiv’s requirements rather than use the archive as an unlimited channel for posting every intermediate text.

The main practical conclusion is simple: before submitting, authors need to check not only grammar and formatting but also the independence of the result, transparency about AI use, data accuracy, and the readiness of all coauthors. The temporary quota limits the number of attempts, while responsibility for the content, errors, and misleading materials remains with the authors.

Frequently asked questions

Does a rejected paper count toward the monthly limit?

According to the rule’s description, yes. The limit applies to submissions, so a rejected paper uses one of the two monthly attempts.

Does the quota apply across different scientific categories?

Yes, if the description in the arXiv announcement applies without exceptions. All new submissions by one submitter are counted together, even if they belong to different arXiv categories.

What happens to a submission that remains under review?

It does not reduce the following month’s limit if it belongs to the previous month, but it counts toward the total number of active submissions. No more than three submissions may remain active at the same time.

Can a paper be submitted on behalf of coauthors?

Yes, if the submitter has obtained the necessary permission in advance. In that case, the monthly attempt is deducted from the submitter, while the coauthors must agree to the submission and the publication details.