arXiv Limits Submissions to 2 Per Month After Record 40,363
arXiv now limits authors to 2 submissions per month after receiving a record 40,363 papers in September 2026, driven by AI-generated content.
👁 115 reads
arXiv, the world’s largest preprint repository, now limits individual authors to two submissions per month starting October 1, 2026. The move comes after a record 40,363 submissions in September, driven by AI-assisted papers.
arXiv announced the new policy on its blog on October 1. The temporary measure aims to buy time for better moderation tools and establish guidelines for authors using AI. Mathematician Terence Tao discussed the issue on his blog.
Submission Numbers That Forced arXiv to Act
| September | Number of Submissions |
|---|---|
| 2016 | 9,869 |
| 2024 | 20,569 |
| 2026 | 40,363 |
- Submissions doubled in two years.
- The cs.AI category grew more than sixfold.
- September 2026 alone generated nearly 9,000 support requests.
Moderators - reportedly around 300 volunteers according to media reports - are seeing increasing numbers of thin papers with narrow scope, dense AI-written text, and “salami” papers where one study is sliced into multiple publications. arXiv states explicitly that AI tools make it easier for authors to flood repositories with low-value work.
New Submission Rules
- Maximum 2 submissions per calendar month per person, across all categories combined.
- Maximum 3 active (not yet published) submissions at once.
- Rejected papers count toward the limit.
- Papers withdrawn by the author before announcement don’t count.
- The limit applies only to the person who clicks “submit,” not co-authors.
The limit follows a May 2026 policy that gives authors a one-year publication ban for undisclosed AI-generated content.
arXiv submission rates have steadily increased since their founding in 1991. In September, arXiv received 40,363 submissions in a single month: a new record!
- Cliff Pickover (@pickover) October 3, 2026
Debate in the Research Community
Reactions are divided. Some researchers praise the limit for forcing focus on best work. Critics, including Nathan Lambert, argue it will hurt legitimate research and slow knowledge sharing. Others point to an obvious loophole: since the limit applies to whoever clicks “submit,” a four-person team can submit eight papers monthly by taking turns at the keyboard. In comments on Terence Tao’s blog, one apt observation emerged: the limit doesn’t reduce the number of papers written, it just delays their appearance where everyone looks for them.
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →What This Means for You
If you read AI research, and most of it lands on arXiv first, remember that preprints are unreviewed. With this flood of submissions, the percentage of weak and AI-generated texts is growing. Before believing a headline claiming “scientists proved,” check whether the paper passed peer review and who wrote it. If you publish research with AI assistance, disclose it - hiding it risks a year-long ban. For more on how models fabricate sources and citations, see our test of AI hallucinations across 10 traps, and for how AI genuinely helps in research, check out the case of Claude and a new enzyme system.
Sources: arXiv blog (Oct 1, 2026), Terence Tao’s blog, Cybernews, Startup Fortune. Verified October 4, 2026.
More on this from QUANTSLANT
news·AI Torture Chamber on GitHub: 4M views and the pain debateA GitHub project injected pain signals into small language models live. The call to report it hit 4 million views on X. What did researchers actually find?