By ByteBot
arXiv, the free preprint server that has been the primary distribution channel for computer science and AI research since 1991, now limits every author to two submissions per month. The policy took effect on October 1, 2026. The cause: AI-generated papers have doubled arXiv’s monthly submission volume in two years, from around 20,000 to a record 40,363 in September 2026 alone. The field that benefits most from open research infrastructure has become its biggest threat.
What Changed
The new rate limit is straightforward: two submissions per calendar month, with no more than three active submissions at any given time. The cap applies to whoever hits “submit” — co-authors are unaffected, so a four-person team could technically push eight papers a month if each takes a turn as primary submitter. The rules apply universally across all arXiv categories, not just computer science.
The detail that stings most: rejected submissions still count against your monthly quota. If your paper gets sent back with a moderation flag, you have lost one of your two shots for the month. There are no exceptions for conference deadlines, no expedited approval lanes, and no carryover from unused prior months. arXiv’s updated rate limit policy is explicit on all of these points.
The Numbers That Made This Inevitable
The growth data explains why arXiv ran out of other options. September 2016 saw 9,869 submissions. September 2024: 20,569. September 2026: 40,363 — a new record, and roughly double the volume from just two years earlier. The cs.AI category grew sixfold between 2024 and 2026. That September surge generated approximately 9,000 support tickets for a moderation team that relies heavily on unpaid volunteers. Rejection rates climbed from around 4 percent to between 10 and 12 percent.
The problem is not just volume. Moderators report a sharp rise in “salami papers” — single research contributions fragmented into multiple smaller submissions to inflate publication counts — and AI-generated survey articles that amount to little more than annotated bibliographies. These submissions consume disproportionate moderator time while delivering minimal scientific value. The platform first responded in May 2026 with one-year bans for papers containing hallucinated references or leftover LLM instructions. The arXiv submission limit is the escalation.
The Debate Is Real
The research community has split along predictable lines. Mathematician Alex Kontorovich at Rutgers put the pro-cap case bluntly: “I’m not going to read your 42 papers this month. Pick two, your very best two.” The quality-gating argument holds that submission velocity was never a virtue, and that forcing prioritization improves the signal-to-noise ratio for everyone downstream.
Scott Kominers, a Harvard Business School professor and nearly twenty-year arXiv contributor, takes the opposite view: “arXiv would do better to economise on moderation — not on scholarship.” His argument is that the cap divides the academic community by punishing all researchers for the behavior of a small population of bad actors. Prachee Avasthi framed it more sharply: “How much science enters the record each month is now set by human bandwidth instead of by how much science is being done.”
Both sides are right about different things. Kontorovich is correct that a flood of low-quality papers degrades the platform for everyone. Kominers is correct that a blanket rate limit is the wrong instrument — it hits the legitimate solo researcher and the grad student juggling a conference deadline just as hard as it hits the lab running automated paper generation pipelines.
What Researchers Need to Do Now
If you publish on arXiv, the practical changes are immediate. Budget your two monthly slots before you have papers ready, not after. Prioritize ruthlessly: a companion paper, an ablation study, or a correction now competes with your primary submission for a finite monthly allocation. If you are at a lab with multiple active submitters, coordinate who files as the primary author to distribute the load across the team. And if a paper is borderline ready, hold it — a rejection still consumes a slot.
The Structural Problem This Does Not Solve
arXiv became an independent nonprofit in July 2026, partly to build the funding base needed for better moderation infrastructure. The rate cap buys time for that work to happen, but it is not the solution. Automated provenance checking — verifying that submitted papers have actual data trails, code, and version histories — is what can scale with submission volume. Human moderator bandwidth cannot. The cap acknowledges the problem; it does not fix it.
The recursive irony here is hard to ignore. AI tools made it trivially easy to generate convincing-looking research papers at scale. Those papers flooded the platform that got spammed with AI-generated research papers and distributes AI research to the world. The resulting cap now throttles the legitimate researchers trying to advance that same field. AI has become a liability for its own scientific infrastructure. That is not a sustainable position, and the two-paper limit is only the most visible symptom.
