arXiv is tightening enforcement against unchecked AI-generated scientific papers, warning authors of bans and stricter submission rules amid rising concerns over fabricated citations and unreliable LLM-generated research.
arXiv is aggressively tightening its policies against careless or unchecked AI-generated scientific papers, introducing a strict “one-strike” enforcement approach aimed at protecting the credibility of open scientific publishing.
The open-access research repository warned that authors submitting papers containing obvious unverified large language model (LLM)-generated content could face a one-year ban from arXiv. Violators may also be required to have future submissions accepted by a reputable peer-reviewed venue before being allowed back onto the platform.
According to Thomas Dietterich, “If a submission contains incontrovertible evidence that the authors did not check the results of LLM generation, this means we can’t trust anything in the paper.”
Dietterich said evidence of AI misuse may include “hallucinated references, comments addressed to or generated by the LLM, and other obvious signs of unverified AI-generated content.”
The platform stressed that authors retain “full responsibility” for submitted material regardless of whether AI tools were used. Researchers remain accountable for plagiarised material, biased content, misleading claims, incorrect references, and factual errors copied from LLMs.
New enforcement measures include endorsement requirements for first-time authors, moderator review, section chair confirmation before penalties are imposed, and an appeals process.
Dietterich also described the policy as “a one-strike rule.” The crackdown comes amid growing concern across scientific publishing over fabricated citations and AI-generated misinformation increasingly appearing in academic and biomedical research papers.
