ArXiv, one of the world’s most influential preprint servers, is moving from warnings to punishments in its battle against low‑quality, AI-generated research, introducing a one‑year ban for authors who submit papers showing clear signs of unchecked chatbot output.
Early concerns and groundwork
Over the past few years, ArXiv has faced a surge of what critics call “AI slop” – papers padded with large language model (LLM) text, complete with fabricated references and boilerplate instructions left in by mistake.1 The platform had already tightened entry by requiring first‑time posters to secure endorsements from established authors and limiting certain review and position papers to those already accepted at conferences or journals.2
The new policy takes shape
In mid-May 2026, computer science section chair Thomas Dietterich publicly outlined a stricter regime: if a submission contains “incontrovertible evidence that the authors did not check the results of LLM generation,” ArXiv “can’t trust anything in the paper,” and the authors face “a 1-year ban from arXiv.”2 Examples include “hallucinated references” and meta‑comments like “fill it in with the real numbers from your experiments,” left directly in the manuscript.1
After the ban, any new submissions by those authors must first be accepted at “a reputable peer-reviewed venue” before appearing on ArXiv, effectively putting them on probation.1 Moderators must document problems, and section chairs must confirm them before penalties are imposed; authors can appeal, and cases are limited to “incontrovertible evidence.”12
Accountability, not a ban on AI
Coverage from outlets including Ars Technica, The Verge, TechCrunch, and The Next Web emphasizes that this is not a blanket prohibition on using AI tools.34 Instead, Dietterich stresses that by signing a paper, “each author takes full responsibility for all its contents, irrespective of how the contents were generated.”12 Researchers may still use LLMs for drafting or editing, but if they paste in “inappropriate language, plagiarized content, biased content, errors, mistakes, incorrect references, or misleading content” without checking, they alone are accountable.2
Why it matters
ArXiv’s move comes as studies show fabricated citations are rising sharply across the scientific literature, raising fears that unvetted AI output could contaminate the record at scale.4 Because ArXiv preprints are widely read and cited before formal peer review, its new sanction is one of the clearest signals yet that major platforms will punish careless use of generative AI in science.4