arXiv to Impose Year-Long Bans on Researchers Who Submit ‘AI Slop’
Table of Contents
The War on ‘AI Slop’ in Academia
arXiv, the indispensable open-access repository for preprints in computer science, physics, and mathematics, is escalating its fight against the influx of low-quality, AI-generated research. In a move designed to protect the integrity of the scientific record, the platform is introducing a strict penalty: authors who submit papers containing “incontrovertible evidence” of unchecked LLM generation will face a one-year ban from the site.
For decades, arXiv has served as the primary engine for rapid information exchange in STEM fields, allowing researchers to share findings before the often-glacial pace of formal peer review. However, the rise of large language models (LLMs) has introduced a new variable into the ecosystem—what some are calling “AI slop.” This refers to papers that may look structurally sound but are riddled with fabrications, logical gaps, and the telltale hallmarks of generative AI.
Defining ‘Incontrovertible Evidence’
The crackdown isn’t a blanket ban on using AI tools. Instead, the focus is on the failure of human oversight. Thomas Dietterich, chair of arXiv’s computer science section, clarified that the goal is to ensure authors take “full responsibility” for every word and claim in their submissions, regardless of how the text was produced.
The “smoking gun” that will trigger a ban includes blatant LLM artifacts. Dietterich pointed to hallucinated references—citations for papers that do not exist—and the accidental inclusion of prompt-and-response text, such as “As an AI language model, I cannot…” or remnants of a conversation with a chatbot. When these errors make it into a final submission, the platform concludes that the authors didn’t even bother to read their own work.
“If a submission contains incontrovertible evidence that the authors did not check the results of LLM generation, this means we can’t trust anything in the paper,” Dietterich stated. This lack of trust transforms a scientific contribution into a liability for the community.
The Cost of a One-Strike Rule
The penalty for these lapses is severe. A one-year ban is a significant blow for researchers, particularly those in fast-moving fields where being the first to post a preprint can establish priority for a discovery. Furthermore, once the ban expires, the authors will be subject to a stricter vetting process: any subsequent submissions must first be accepted by a reputable, peer-reviewed venue before they can be uploaded to arXiv.
To prevent arbitrary censorship or mistakes, the process involves a multi-step verification. Moderators must first flag the suspicious content, and section chairs must confirm the evidence before the ban is enacted. There is also a formal appeal process for authors who believe a mistake has been made.
A Systemic Crisis of Fabrication
This move comes as the broader scientific community grapples with a surge in fabricated data. Recent studies in biomedical research have highlighted a rising trend of fake citations, a phenomenon heavily attributed to the seductive ease of using LLMs to “fill in the blanks” of a literature review.
arXiv is also undergoing a structural evolution to better handle these challenges. After more than 20 years under the umbrella of Cornell University, the organization is transitioning into an independent nonprofit. This shift is expected to provide the financial and operational flexibility needed to implement more robust moderation tools and staffing to combat the growing volume of automated submissions.
While arXiv has already implemented an endorsement system for first-time posters to curb spam, the new ban signals that the platform is no longer treating AI-generated errors as simple mistakes, but as a breach of academic trust.
