ArXiv has announced tougher rules on research papers generated by AI, warning that authors who let AI do all the work without proper quality and accuracy checks will face a one‑year ban from submitting to the repository. To submit again after the one-year penalty, preprints must have been accepted by another reputable platform.
ArXiv is an open-access repository where researchers publish “preprints,” or early versions of research papers that have not yet undergone peer review. First-time submitters already need to be endorsed by an established arXiv author in order to publish their work, but with the proliferation of generative AI tools, the platform is working overtime to ward off low-quality (at best) and artificial (at worst) papers.
The repository’s computer science chair, Thomas Dietterich, says authors are fully responsible for everything in their manuscripts, including content generated by LLMs. Mods will consider papers untrustworthy if they find clear evidence that the authors did not check the output generated by AI, like hallucinated references, falsified data, leftover AI meta‑comments or prompts, or chatbot responses (e.g., “Do you want me to make it sound more professional?”)
This new rule adds a human peer‑review gate on top of arXiv’s usual role as an open preprint server. Dietterich says authors will have the opportunity to appeal decisions, but the bar for overturning a ban will likely be high.
It’s important to note that this policy does not completely prohibit researchers from using AI tools. ArXiv allows researchers to use LLMs for tasks such as drafting and editing, but they must review the output, correct any mistakes, and report their use in accordance with the platform’s guidelines.
