arXiv has applied the oldest tool in API management to scholarly publishing. In Fair Moderation, Equitable Access, and AI: arXiv’s Updated Rate Limit Policy, the preprint server limits every submitter, from October 1, to “two submissions per calendar month, with a limit of three total active submissions at any given time,” across all categories. The reason is stated without hedging: before AI, there was “an easily discernible, practical limit to the rate at which independent submissions could be produced,” and now that limit “is up for debate.” The policy is called “a stopgap while we determine what may be the new best practice.”
The figures are arXiv’s own and they are stark. In September 2016 arXiv received 9,869 submissions. In September 2024 it received 20,569. This September it received 40,363, a record, which “generated almost 9,000 support tickets for arXiv staff and moderators.” Submissions doubled in two years, and the computer science AI category grew more than six times. arXiv’s policy allows AI as a disclosed research tool, but moderators are seeing “thin papers of narrow scope” and “salami” papers split into many submissions, from “a relatively small proportion of authors.” The design choices are the ones a good rate limit makes: the limit counts submissions, not announced papers, “because it is submissions that consume moderator time”; it applies to the submitter, not co-authors; held papers do not count against the next month; and a paper deleted before announcement does not count at all.
The catalog holds arXiv as a provider of reading, not writing. The arXiv provider page lists 4 API pages, the arXiv Query API, the arXiv OAI-PMH API, arXiv Bulk Data and arXiv RSS Feeds, all of them for retrieving the corpus this policy protects. The submission path has no public API on the record. The agentic access profile is read-only. Spec presence, agentic access, OpenAPI examples and the rate limit signal are lit on the Agent Readiness score.
The Kin Score is 45.4, developing band, with discoverability at 73.2 and contract quality at 58.4 against access clarity at 26.3. The Agent Readiness score is 16.6, agent-aware. The rate limit signal is the one agent dimension that maps directly to this story, and it is lit for the read side. The policy itself is the more interesting artifact: a published limit, with the unit chosen to match the scarce resource, exceptions written out in a FAQ, and a stated intention to revisit it. That is what a good API rate limit document looks like, written for a human submission form. arXiv has rate-limited the input to a scarce human process. Every provider whose agent traffic is outgrowing its reviewers will be writing the same page.