arXiv bans low‑quality AI papers

arXiv has not announced a blanket ban on all AI-assisted papers. What its published rules do show is a tighter moderation framework that gives staff and volunteer moderators more room to reject low-quality submissions, including machine-generated ones that fail basic scholarly standards. arXiv’s submission guidance says papers must be “topical and refereeable scientific contributions” that follow accepted standards of scholarly communication, while its moderation policy says submissions can be declined if they lack originality, novelty, significance, or careful preparation. (info.arxiv.org 1) (info.arxiv.org 2) That matters because arXiv sits upstream of formal peer review for much of computer science, physics and mathematics. (info.arxiv.org) The repository says all submissions undergo moderation before public posting, even though that process is not peer review. ### So what actually changed? The clearest public signal is not a standalone “AI paper ban” notice but the way arXiv now links generative-AI use to its moderation and conduct framework. A January 2023 arXiv policy said authors could use generative AI tools to help prepare manuscripts, but those tools could not be listed as authors and authors remained responsible for the content. That blog post also directed readers to the moderation page for the current policy. (info.arxiv.org) The current moderation page is broader and more operational. It says submissions must be self-contained, of scholarly archival interest, and prepared with appropriate sections, figures, tables and references. (info.arxiv.org) It also says arXiv may decline work that lacks originality or contains serious misrepresentations. In practice, that gives moderators a basis to screen out low-effort AI-generated manuscripts that look scientific but do not meet those thresholds. That is an inference from the published standards rather than wording arXiv spells out in a single sentence. ### Is this a one-year ban? arXiv’s public enforcement page says users who violate policy can have submission privileges suspended “for a designated period,” but the page does not specify a standard one-year term for AI-generated papers. (blog.arxiv.org) It lists possible account-level actions including identity checks, flagged accounts, suspension of submission privileges and full account suspension. So the strongest verified version of the story is narrower than the social-media framing: arXiv has published enforcement tools that can temporarily block submitters, but the exact duration appears to be discretionary unless arXiv states otherwise in a separate notice not surfaced in its public help pages. (info.arxiv.org) ### What counts as “low-quality” under arXiv’s rules? arXiv’s moderation policy gives a concrete checklist. Submissions can be declined for failing scholarly standards in form, for lacking “general scrupulousness and care of preparation,” for missing originality or significance, or for including falsified, plagiarized or misrepresented content. (info.arxiv.org) It also says non-substantive material such as course projects, proposals, news and political advocacy may be declined. For AI-generated papers, that means the issue is less whether a language model touched the draft and more whether the final manuscript is coherent, attributable and scientifically substantive. arXiv’s metadata rules separately say generative AI language tools should not be listed as authors. (info.arxiv.org) ### Why does reproducibility keep coming up? arXiv’s submission system already pushes authors toward portable source files, complete figures and self-contained materials. The platform says omitted figures are not accepted, and it encourages inclusion of ancillary files such as data and programs. Those are archival and discoverability rules, but they also align with the broader demand for reproducible research workflows. (info.arxiv.org) The moderation page does not use “reproducibility” as a headline enforcement term. But it does require careful preparation and substantive research content, which are the same fault lines critics point to when low-effort AI-generated papers flood preprint systems. (info.arxiv.org) That connection is again an inference from the published policy language. ### What should authors watch next? arXiv says quality-assurance checks can take one to four days or longer, and authors are notified by email if problems are found. It also says declined submissions can be appealed, though in some cases arXiv may require journal acceptance before considering an appeal. (info.arxiv.org) The practical next step is to watch arXiv’s moderation, submission and policy pages for any explicit AI-specific update that goes beyond the current standards-and-enforcement framework. I found clear evidence of stricter moderation tools and author responsibility for AI-assisted content, but I did not find a public arXiv page that explicitly states a universal one-year ban for low-quality AI-generated papers. (info.arxiv.org) (info.arxiv.org)