IthacaArXiv bans AI-slop submitters for one year.
The repository underpins physics, math, and machine learning research; it is now forced into quality-filtering because journals haven’t policed AI-generated work.
Banned authors must subsequently publish at a peer-reviewed venue before resubmitting, computer science chair Thomas Dietterich said.
Sources: 404 Media
WHAT ARXIV ACTUALLY IS
arXiv is not a journal. It is a preprint server founded at Los Alamos in 1991 by physicist Paul Ginsparg, designed to circulate papers *before* peer review. The entire culture of physics, math, and now machine learning shifted to posting first and submitting to journals later — or never.
THE MODERATION MODEL
arXiv has always relied on volunteer moderators — working researchers who screen submissions for basic legitimacy, not correctness. The endorsement system requires new authors to be vouched for by established ones. This was sufficient when the cost of writing a fake paper was a graduate student's month; LLMs collapsed that cost to minutes.
WHY JOURNALS FAILED FIRST
Peer review is unpaid labor done by overworked academics. A reviewer who suspects AI generation has no efficient way to prove it — detection tools are unreliable, and editors are reluctant to reject on suspicion alone. The pipeline that arXiv assumed would catch slop downstream simply isn't catching it.
THE INVERTED GATEKEEPER
arXiv now demands prior peer-reviewed publication before reinstatement — the opposite of its founding premise, which was to *escape* journal gatekeeping. A preprint server is being forced to act like a journal because journals stopped acting like journals.
THE SCALE PROBLEM
arXiv now receives over 20,000 submissions per month, with computer science the fastest-growing category. Volunteer moderation scales linearly; LLM-generated submissions scale with compute. The arithmetic does not work, which is why bans — a blunt instrument — have replaced screening.
THE GINSPARG WARNING
Ginsparg himself wrote a 2021 essay warning that arXiv's openness was being exploited by predatory authors padding citation counts and CVs. The current ban is the institutional acknowledgement of a problem the founder named four years before the moderation system formally cracked.