As a preprint platform widely used by researchers globally, arXiv has recently tightened its management policies regarding AI-generated papers. Long an important channel for disseminating papers in fields such as computer science and mathematics, arXiv sees its content rapidly influence academic discourse and research directions despite submissions lacking peer review at the time of upload. Previously, arXiv had already begun taking measures to address the growing number of low-quality AI papers. For instance, first-time submitters must now obtain endorsement recommendations from senior researchers. Furthermore, after more than two decades under the stewardship of Cornell University, arXiv is transitioning into an independent non-profit organization to secure additional funding for addressing issues like AI misuse. The latest rules were announced by Thomas Dietterich, head of the Computer Science section at arXiv. He stated that if "clear evidence" indicates authors failed to verify results generated by large models, the credibility of the entire paper will be questioned. Such evidence includes "fabricated references," directly retained AI prompts, or dialogue content with large language models. Under the new regulations, once these issues are confirmed, authors will be banned from submitting to arXiv for one year. Subsequently, their follow-up manuscripts must first be accepted by formal peer-reviewed journals or conferences before being uploaded again to arXiv. However, this does not constitute a blanket ban on using large models. arXiv emphasizes that researchers may still utilize Large Language Models (LLMs) for writing assistance but remain fully responsible for all content in their papers, regardless of whether it was generated by AI. In media interviews, Thomas Dietterich noted that while this represents a "one-strike-and-you're-out" policy, enforcement requires prior flagging by moderators, confirmation by section heads, and allows authors the right to appeal. In recent years, concerns within academia over AI hallucinations have continued to escalate. Peer-reviewed studies indicate that the number of "fabricated citations" in biomedical literature is rising, with large language models identified as a significant contributing factor.