arXiv Had to Cap Submissions. I Read Why.
AI-assisted writing doubled the volume to 40,000 submissions a month. The cap is a symptom, not the disease.
arXiv now limits submitters to two papers a month after September 2026 hit 40,363 submissions, nearly double two years prior. The stated driver is AI-assisted writing flooding the intake. The cap is a bandage on a much larger shift: the scientific record's ability to absorb generated text has a limit, and we've hit it.
The uncomfortable part is that the cap doesn't distinguish good from generated. A human with two real papers and a lab spamming generated submissions both hit the same wall. It's a blunt instrument, and blunt instruments are what you reach for when the volume breaks the system faster than the curation can.
For researchers, the signal is clearer than the policy: the bar for a submission that matters just rose. If your paper reads like it was written by a model and adds nothing, it's now competing with forty thousand others for attention it won't get. The cap is arXiv's problem; the noise is everyone's.
I read the news and changed my own habit. I write the draft myself and use models only to check, not to compose. The cap is a reminder that generated prose is now abundant and cheap, which makes the human-authored, actually-novel paper the scarce thing.