ArXiv Bans Researchers for Unverified AI-Generated Content in Papers

ArXiv, the preeminent open-access repository for scientific preprints, has announced a new policy banning researchers who submit papers containing unverified AI-generated content, including hallucinated references and unedited large language model (LLM) outputs, according to a report from The Verge. The policy, effective immediately, targets submissions with ‘incontrovertible evidence that the authors did not check the results of LLM generation,’ as stated in the platform’s updated guidelines.

The platform, hosted at Cornell University and widely used by U.S. researchers, will now bar repeat offenders from uploading papers to the system. Violations include leaving ‘meta-comments’ from AI tools in final drafts or citing non-existent sources generated by LLMs. ‘This is about maintaining trust in scholarly communication,’ an ArXiv spokesperson said in the Verge report. ‘Unverified AI content undermines the integrity of the scientific record.’

The policy reflects growing concerns about ‘AI slop’—low-quality or unvetted AI-generated content—in academic publishing. While AI tools are increasingly used for drafting and data analysis, critics argue many researchers fail to verify outputs, leading to errors and fabricated citations. The move positions ArXiv as one of the first major academic platforms to enforce strict accountability for AI use in scholarly work.

ArXiv, which hosts over 2 million preprints annually, serves as a critical resource for researchers in physics, computer science, and mathematics. The policy change comes as U.S. institutions and journals grapple with establishing standards for AI-assisted research. Similar initiatives are being considered by the American Psychological Association and the National Science Foundation.