ArXiv Introduces One-Year Ban for Researchers Submitting Papers with Unchecked AI-Generated Content
May 17, 2026 – 1:37 pm
TL;DR
ArXiv will ban researchers for one year if they submit papers with obvious signs of unchecked AI generation, such as hallucinated references or leftover chatbot instructions. This policy, announced by computer science section chair Thomas Dietterich, is the first formal penalty by a major preprint platform for AI-generated content deemed unacceptable.
ArXiv, the open-access repository that has long served as the primary distribution channel for preprint research in computer science, mathematics, and physics, will impose a one-year ban on authors who submit papers containing "incontrovertible evidence" of unvetted large language model output. Dietterich stated that such submissions indicate "we can’t trust anything in the paper."
Key Points:
- The policy does not ban the use of AI tools entirely; researchers can still utilize language models for drafting, editing, or analysis.
- Penalties are triggered by evidence that an author pasted LLM output into a paper without checking it, leading to issues like hallucinated references or fabricated data tables with placeholder instructions.
- If moderators identify such evidence and a section chair confirms it, the author faces a one-year ban from arXiv, after which all subsequent submissions must be accepted by a peer-reviewed journal before appearing on the platform.
Why It Matters:
ArXiv’s quality standards are unusually consequential because papers posted there are frequently cited and built upon before they appear in formal publications. A hallucinated citation on ArXiv can spread through the research literature just as effectively as one in a peer-reviewed journal.
A study published in The Lancet in May 2026 found that fabricated citations have increased significantly since 2023, attributed to the proliferation of AI writing tools. Roughly one in 458 papers contained at least one fake reference by 2025, rising to one in 277 in the first seven weeks of 2026.
ArXiv has a compelling reason to address this issue.