ArXiv to Ban Authors for One Year if They Submit AI-Generated Papers Without Human Review

BitcoinWorld

ArXiv to Ban Authors for One Year if They Submit AI-Generated Papers Without Human Review

ArXiv, the widely used open-access repository for preprint research, has announced a new policy that could result in a one-year ban for authors who submit papers containing clear evidence of unchecked AI-generated content. The move, outlined Thursday by Thomas Dietterich, chair of ArXiv’s computer science section, targets the growing problem of low-quality, AI-produced research that undermines trust in scientific publishing.

What the New Rule Means for Researchers

Under the updated guidelines, if moderators find ‘incontrovertible evidence’ that authors did not verify the output of large language models (LLMs) before submission, the paper will be rejected and the authors face a one-year suspension from posting on ArXiv. After the ban, authors must have subsequent submissions accepted by a reputable peer-reviewed venue before returning to the platform.

Dietterich specified that such evidence includes fabricated references, nonsensical citations, or direct copy-paste errors from an LLM. The policy does not prohibit the use of AI tools entirely; rather, it holds authors ‘fully responsible’ for all content, regardless of how it was generated. This includes plagiarism, biased statements, and factual inaccuracies introduced by AI.

Why This Matters for Scientific Integrity

ArXiv has long been a cornerstone of rapid research dissemination, especially in computer science, mathematics, and physics. However, the rise of generative AI has led to a surge in submissions that appear to be produced with minimal human oversight. Recent peer-reviewed studies have documented an increase in fabricated citations in biomedical literature, likely linked to LLM use.

By enforcing this one-strike rule, ArXiv aims to preserve the credibility of its repository. The policy also includes an appeals process, allowing authors to contest decisions. Moderators must first flag issues, and section chairs must confirm evidence before penalties are applied.

Broader Implications for the Research Community

This policy reflects a growing consensus across academia that AI tools should assist, not replace, human oversight in research. ArXiv’s transition to an independent nonprofit organization, after being hosted by Cornell University for over two decades, gives it more flexibility to enforce such measures. The repository has already taken steps to curb AI-generated submissions, including requiring endorsements for first-time posters.

For researchers, the message is clear: using AI to draft or polish language is acceptable, but submitting work without rigorous fact-checking and citation verification is not. This aligns with broader editorial standards in scientific publishing, where accountability remains paramount.

Conclusion

ArXiv’s new ban policy represents a significant step in maintaining the integrity of preprint research in an era of widespread AI use. By penalizing authors who fail to review AI-generated content, the repository reinforces the principle that human researchers bear ultimate responsibility for their work. As AI tools become more integrated into the research process, such guardrails will likely become standard across academic publishing.

FAQs

Q1: Does ArXiv’s new policy ban the use of AI in writing papers?
No, it does not ban AI use. It bans the submission of papers with clear evidence that authors did not check AI-generated content for errors, such as fabricated references or nonsensical text.

Q2: What counts as ‘incontrovertible evidence’ of AI misuse?
Examples include hallucinated citations, references to nonexistent sources, and direct copy-paste errors from an LLM that indicate no human review took place.

Q3: Can authors appeal a ban?
Yes, the policy includes an appeals process. Moderators must flag the issue, section chairs confirm the evidence, and authors can contest the decision.

This post ArXiv to Ban Authors for One Year if They Submit AI-Generated Papers Without Human Review first appeared on BitcoinWorld.