**ArXiv Limits AI-Generated Review Papers in Computer Science Category Amid Surge in Submissions**

ArXiv, the free research repository founded at Cornell University and widely used by scientists and technologists around the world, has announced a significant policy change for its Computer Science category. Starting immediately, the platform will no longer accept review articles or position papers unless they have already undergone peer review and been accepted at a recognized journal or conference.

The policy shift, announced on October 31, comes in response to a dramatic increase in AI-generated survey papers, which moderators describe as “little more than annotated bibliographies.” Whereas historically ArXiv received a small number of high-quality review articles—typically authored by senior researchers—the repository now faces hundreds of these submissions every month.

An official statement on ArXiv explained the situation: “In the past few years, ArXiv has been flooded with papers. Generative AI/large language models have added to this flood by making papers—especially papers not introducing new research results—fast and easy to write.”

Thomas G. Dietterich, an ArXiv moderator and former president of the Association for the Advancement of Artificial Intelligence, elaborated on X (formerly Twitter): “We were driven to this decision by a big increase in LLM-assisted survey papers. We don’t have the moderator resources to examine these submissions and identify the good surveys from the bad ones.”

The rise of AI-assisted research writing is supported by recent studies. Research published in *Nature Human Behaviour* found that nearly a quarter of all computer science paper abstracts showed signs of large language model modification as of September 2024. Similarly, a study in *Science Advances* reported a sharp increase in AI use in research papers published in 2024 following the launch of ChatGPT.

ArXiv relies on volunteer moderators who filter submissions based on scholarly value and topical relevance but do not perform peer reviews. While review articles and position papers were never officially accepted content types, moderators previously allowed exceptions for work from established researchers or scientific societies. However, this discretionary system has become unsustainable given the volume of AI-generated content.

The research community’s reaction to the new policy has been mixed. Some, like AI safety researcher Stephen Casper, worry it may unfairly impact early-career researchers and those working on ethics and governance—groups that frequently produce review and position papers. “Review/position papers are disproportionately written by young people, people without access to lots of compute, and people who are not at institutions that have lots of publishing experience,” Casper noted in his critique.

Others have sharply criticized ArXiv’s decision as misguided or shortsighted, while some support stricter measures, including the use of AI detection tools to identify generated content. However, AI detection tools remain unreliable and prone to high false-positive rates, risking wrongful rejection of legitimate research. Adding to the complexity, a recent study found that researchers failed to detect one-third of ChatGPT-generated medical abstracts as machine-written. Furthermore, the American Association for Cancer Research reported that fewer than 25% of authors disclosed AI use in their papers, despite mandatory disclosure policies.

Under the new ArXiv policy, authors wishing to submit review or position papers to the Computer Science category must provide documentation of successful peer review, including journal references and DOIs. Reviews from workshops will not satisfy this standard.

Importantly, ArXiv clarified that this policy change currently applies only to the Computer Science section, though similar measures may extend to other categories if they experience comparable surges in AI-generated submissions.

—

*This update reflects ArXiv’s efforts to maintain the quality and integrity of its repository amid rapidly evolving challenges posed by AI-assisted research writing.*
https://decrypt.co/347196/arxiv-blocks-ai-generated-survey-papers-flood-trashy-submissions