Between October 2025 and February 2026, arXiv tightened three submission policies: Computer Science surveys and position papers now require documented prior peer review, new-author endorsement can no longer rest on an institutional email address alone, and every submission needs a full English-language version. Machine-generated translation is explicitly permitted.
Why arXiv moved on three fronts at once
The timing is not a coincidence. arXiv also completed its own governance transition on July 1, 2026, leaving Cornell University’s institutional umbrella to become arXiv, Inc., an independent 501(c)(3) nonprofit (see CASRAI’s coverage of the arXiv spinout for what changed there). A smaller, resource-constrained volunteer moderation corps facing a rising volume of low-quality and AI-assisted submissions is the throughline across all three policy changes below: each one is arXiv trying to protect scarce moderator time rather than expand it.
Endorsement: institutional email is no longer enough on its own
arXiv has long used an endorsement system to gate first-time submissions in a given subject area: a new author normally needs an existing arXiv author in that category to vouch for them, though arXiv has historically also accepted a recognized academic or research-institution email address as an alternative, lower-friction path to that first submission.
As of arXiv’s January 21, 2026 policy update, an institutional email address is no longer sufficient by itself to qualify for endorsement arXiv-wide. arXiv first piloted the tightened rule in the Mathematics category in December 2025 before extending it to all categories the following month. arXiv has attributed the change to an unsustainable rise in non-scientific and low-quality submissions arriving via the institutional-email path, which was driving up rejection rates and consuming moderator and staff time that arXiv describes as no longer sustainable at the prior volume.
For research administrators supporting first-time preprint authors — postdocs, incoming faculty, or graduate students without an established co-author network on arXiv — this means budgeting more time for a genuine peer endorsement rather than assuming an institutional email address will clear the bar on its own.
English-language requirement: full translations, not just abstracts
Announced on November 21, 2025 and effective February 11, 2026, arXiv now requires that every submission include a full English-language version of the paper, not merely an English-language abstract as was previously sufficient. Authors who write in another language can satisfy the requirement either by submitting the paper originally in English or by including a full English translation alongside the original.
arXiv has been explicit that machine-generated translation is an acceptable route, provided the translated version is faithful to the original paper’s content. arXiv’s stated rationale is primarily moderation capacity: its volunteer moderator pool is not fluent in every submission language, and a full English version lets moderators actually screen a paper’s content rather than relying on an abstract alone — alongside a secondary goal of widening the readership of non-English-language research. The policy drew substantial pushback from parts of the mathematics and physics communities, who argued it disadvantages non-English-speaking authors and sits uneasily against arXiv’s own stated commitment to broadening participation from outside anglophone research systems.
Practically, this is now a submission checklist item alongside formatting and category selection: a paper without a full English-language version, translated or original, will not clear moderation regardless of the abstract’s language.
AI-generated content: prior peer review now required for CS surveys and position papers
Ahead of both of the above, arXiv tightened its Computer Science category moderation in October 2025, requiring that literature-survey and position-paper submissions come with documentation of prior acceptance through peer review — a formal journal or conference review process — before arXiv will list them. Papers on the societal impact of science and technology remain an explicit exception and can still be submitted without that prior peer-review documentation.
arXiv’s moderators described the trigger plainly: a sharp rise in submissions that were, in substance, AI-generated literature surveys — described by moderators as amounting to little more than annotated bibliographies — arriving in volumes moderators could not reasonably screen by hand. This sits alongside arXiv’s broader, longer-standing generative-AI disclosure expectations for authors, which this narrower rule supplements rather than replaces: the October 2025 change targets a specific submission type (CS surveys and position papers) rather than imposing a blanket restriction on AI-generated content across every arXiv category.
The change has divided researchers. Critics argue it disadvantages early-career authors and smaller research groups who may write a genuinely useful survey without first clearing a formal peer-review gate; supporters frame it as a necessary filter against a submission type moderators say became effectively unscreenable at the volume it was arriving.
What this means for researchers and research administrators
- Before a first arXiv submission from a new author: confirm who will provide endorsement in the target category — don’t assume an institutional email address alone will clear it.
- For non-English-language manuscripts: budget time for a full English translation (machine translation is acceptable if faithful to the source), not just an English abstract, before submission.
- For CS literature reviews, surveys, or position papers: secure and document formal peer-review acceptance before submitting to arXiv, or route the work toward a venue that doesn’t carry that prerequisite.
- For research offices supporting preprint policy questions: these three changes did not alter arXiv’s core value proposition — free-to-read, free-to-post preprints with lightweight moderation — but they meaningfully raise the bar for what “lightweight” means at first submission.
For the fundamentals of what arXiv is and how to use it, see CASRAI’s guide to arXiv preprints and the broader preprint servers explained guide. If you’re choosing between preprint servers for a given field, see CASRAI’s comparisons of arXiv vs. bioRxiv, arXiv vs. Research Square, and arXiv vs. SSRN.
Frequently asked questions
Does the English-language requirement mean arXiv no longer accepts non-English papers?
No. Non-English papers are still accepted, but as of February 11, 2026 they must be accompanied by a full English-language version — either the paper submitted originally in English or a complete translation alongside the original — rather than the English abstract that was previously sufficient on its own.
Can I use ChatGPT or another AI tool to produce the required English translation?
arXiv has said machine-generated translations are acceptable, provided the translated content is faithful to the original paper. This is a narrower allowance than a general AI-generated-content policy: it applies to translation of an author’s own paper, not to using AI to generate the paper’s substantive content.
Does the new endorsement rule apply to every arXiv category?
Yes, as of the January 2026 rollout. arXiv first tightened endorsement in the Mathematics category in December 2025, then extended the same institutional-email restriction to all arXiv categories in January 2026.
Does the peer-review-first rule for surveys apply outside Computer Science?
As announced, the requirement that literature surveys and position papers document prior peer-reviewed acceptance is a Computer Science category policy, with an explicit exception for papers on the societal impact of science and technology. Check arXiv’s current category-specific policies before submitting, since moderation policies can vary by category and continue to evolve.
Is any of this connected to arXiv leaving Cornell?
The endorsement, language, and AI-content policy changes were each announced and justified independently, primarily around moderator capacity and submission quality. They land in the same window as arXiv’s July 2026 transition to independent nonprofit status, but arXiv’s own announcements do not present them as directly caused by the governance change — both reflect the same underlying pressure of rising submission volume against a volunteer moderation model.
This page tracks arXiv’s own policy announcements as of the verification date below. arXiv’s submission and moderation policies continue to evolve; consult arXiv’s official help pages for the current requirements before submitting.
