ArXiv’s New Policy Targets Low-Quality Submissions
For over three decades, ArXiv has been the go-to repository for preprint research in computer science, mathematics, physics, and related fields. Its open-access model has accelerated scientific communication, but a recent surge in submissions that appear to be generated by artificial intelligence without human oversight has prompted a clear response. As of March 2025, ArXiv will ban authors for one year if they submit papers that show obvious signs of unchecked AI generation. The new rule, announced by Thomas Dietterich, chair of ArXiv’s computer science section, aims to preserve the integrity of the repository as a trusted source of scholarly work.
Why ArXiv Is Taking Action
The problem of low-quality, AI-produced manuscripts has grown as large language models like ChatGPT and GPT-4 become widely accessible. Some researchers have submitted papers that appear to be hastily assembled by AI tools, with little or no human editing, fact-checking, or even reading. These submissions often contain nonsensical text, hallucinated citations, and irrelevant figures. Dietterich noted that such papers waste reviewers’ time and dilute the value of ArXiv’s preprint database. The one-year ban is intended to deter bulk submissions that violate the repository’s basic standards.
How the Ban Will Work
Under the new policy, if a submission is flagged as containing “obvious signs of unchecked AI generation,” the submitter will receive a warning for a first offense. A second violation within a rolling 12-month period triggers a one-year ban from submitting any new papers. The ban applies to the individual author account, not to co-authors, though repeat offenders may face longer restrictions. ArXiv will rely on both automated detection tools and human moderators to identify problematic papers.
What Counts as Unchecked AI Generation?
The policy defines obvious signs based on common AI pitfalls: meaningless or repetitive phrasing, fake references, hallucinations about well-known results, and inconsistent formatting that deviates from standard academic style. Papers that use AI as a writing aid—such as for grammar checks or idea generation—are not penalized as long as the author has thoroughly reviewed and edited the manuscript. According to Dietterich, “The key is that the human author must be responsible for the content and ensure its accuracy.”
Industry and Academic Reactions
Reactions from the research community have been largely supportive. Many scientists have pointed out that the problem of AI-generated junk papers is not unique to ArXiv; other preprint servers and journals face similar challenges. By taking a firm stance, ArXiv hopes to set a standard for quality control in open-access publishing. Some critics caution that automated detection might flag legitimate uses of AI, but ArXiv has stated it will provide an appeals process for banned authors.