ArXiv Cracks Down on "AI Slop" with One-Year Ban
ArXiv, a leading open-access repository for scientific preprints, has declared a stringent new policy targeting submissions that exhibit clear signs of unchecked artificial intelligence generation. Authors found to have submitted "obviously AI-generated work" will now face a one-year ban from the platform. This decisive action comes as academic journals and preprint servers grapple with an increasing volume of low-quality, AI-assisted content, often referred to as "AI slop."
The policy specifically targets instances where authors fail to adequately verify the output of large language models (LLMs). Thomas Dietterich, chair of ArXiv's computer science division, emphasized that if generative AI tools produce inappropriate language, plagiarized content, biased information, errors, incorrect references, or misleading content, and that output is included in scientific works, the responsibility lies solely with the author(s).
Defining "Incontrovertible Evidence" of Unchecked AI Use
The core of ArXiv's new enforcement lies in identifying "incontrovertible evidence" that authors neglected to review the results of LLM generation. This evidence includes several key indicators:
- Hallucinated references: Citations that are entirely fabricated or do not correspond to real academic papers.
- Meta-comments from the LLM: Unedited prompts or internal notes from the AI, such as "here is a 200-word summary; would you like me to make any changes?" or "the data in this table is illustrative, fill it in with the real numbers from your experiments."
- Placeholder text: Unfilled sections like "REF NEEDED" left within the manuscript.
Dietterich clarified that these examples serve as clear indications that the authors did not adequately check the AI-generated content, leading to a lack of trust in the entire paper. This policy is not an outright ban on using AI tools, but rather a strong reinforcement of author accountability.
Consequences and Path to Reinstatement
The penalty for submitting unverified AI-generated content is a one-year ban from ArXiv. Following this ban, authors wishing to submit to ArXiv again will face an additional requirement: their subsequent submissions must first be accepted at a reputable peer-reviewed venue. This two-tiered approach aims to ensure a higher standard of quality control for returning authors. The decision to impose a ban will involve a moderator documenting the problem, followed by confirmation from the Section Chair, with an appeals process available.
ArXiv's existing Code of Conduct already stipulates that by signing as an author, each individual takes full responsibility for all content, regardless of how it was generated. This new clarification serves to formalize and publicly announce the enforcement of these principles in the context of generative AI.
Backlash and Broader Implications for Research
The implementation of ArXiv's one-year ban has not been without controversy, sparking a significant debate within the research community. While many welcome the move as a necessary step to maintain the integrity of scientific publishing, others express concerns. Some researchers argue that expecting authors to manually check every single reference in papers, especially those with numerous co-authors, might be overly burdensome. There are also concerns about the potential for uneven enforcement and the impact on smaller research teams or startups that might rely more heavily on LLMs for drafting and editing.
The influx of AI-generated submissions has been a growing problem, with ArXiv reportedly forced to shut down its computer science review section in November 2025 due to the overwhelming volume of AI-generated papers, many of which lacked new research results. This policy highlights a broader tension between the speed of scientific dissemination, the increasing accessibility of powerful AI tools, and the fundamental need for verifiable, high-quality research.