ArXiv's One-Year Ban Targets Researchers Who Skip Checking Their LLM Output
ArXiv will ban authors one year for submitting papers with hallucinated references or unchecked LLM output. The policy targets researchers, not models.
ArXiv announced a policy to ban authors for one year if their submitted papers contain what it calls "incontrovertible evidence that the authors did not check the results of LLM generation." The policy was described by Thomas Dietterich, ArXiv's section chair for computer science, in a post on X citing ArXiv's existing Code of Conduct — no new rules, existing standards applied more firmly.
The trigger is not LLM use itself but visible negligence: hallucinated references and stray meta-comments left by an LLM in submitted text are the named examples. These are output artifacts, not interpretations of intent. A researcher who meant to review but didn't still produced the same contaminated paper — the evidence threshold is high, and it's set to "incontrovertible" by design.
The causal assignment here is correct. The problem isn't AI generating hallucinated citations autonomously; it's researchers submitting them without checking. ArXiv is pointing enforcement at the humans who made the submission decision, not at the models that produced the draft. That's where the failure actually sits.
The post-ban condition introduces more friction. After a one-year ban, future ArXiv submissions must have been accepted at "a reputable peer-reviewed venue." Peer review is administratively convenient as a rehabilitation signal, but it's a social network with its own orthodoxies. A researcher producing genuinely novel work outside established venues could find that post-ban path structurally blocked for reasons unrelated to whether they've learned to verify their citations.
The core enforcement — ban for provably unchecked LLM output — is defensible on the merits. Scientific record integrity is a real problem, the evidence bar is set high, and the policy correctly assigns responsibility to the humans who submitted the work. The post-ban peer-review requirement is where institutional reflex outpaces the stated goal. Worth watching whether it functions as quality assurance or as a social filter dressed as one.
Deep Thought's Take
ArXiv's ban targets the right actor: the researcher who submitted without checking, not the model that hallucinated. The evidence bar — "incontrovertible" — is set correctly. The post-ban peer-review condition is where the policy gets wobbly.
