ArXiv, the preprint repository hosting over 2.4 million scientific papers, is implementing stricter enforcement against authors who rely entirely on AI to generate research submissions. The platform will now ban violators for one year, escalating penalties from previous warnings.
The move targets a specific problem: papers where large language models produce the entire content without substantive human input. ArXiv moderators have observed an uptick in low-quality submissions generated primarily by AI systems, which undermine peer review and waste reviewer time. These papers often contain fabricated citations, nonsensical results, or recycled content.
The ban applies to authors who demonstrate no meaningful contribution to their own work. ArXiv distinguishes between legitimate AI use—like running code or drafting sections—and wholesale delegation where humans submit LLM output verbatim. The distinction matters. Researchers using AI as a tool face no penalty. Those treating it as a replacement for scholarly work do.
This enforcement addresses a growing tension in academic publishing. As LLMs become more capable at mimicking academic prose, submission volumes to preprint servers have climbed. Many submissions now fail basic quality checks. ArXiv receives thousands of papers daily, and moderators manually review flagged content.
The one-year ban creates real consequences. For active researchers, being locked out of ArXiv for 12 months damages career visibility and delays knowledge sharing. The penalty forces consideration of academic integrity before hitting submit.
ArXiv's approach differs from commercial platforms. Nature and other journals have issued guidelines for AI disclosure. Some publishers require author statements about AI use. ArXiv goes further by explicitly punishing negligent automation rather than just requiring transparency.
The policy reflects broader academic concern about AI's role in research. Legitimate applications exist: summarizing data, debugging code, drafting introductions. But wholesale paper generation bypasses the intellectual work that separates scholarship from content