ArXiv Draws a Hard Line on AI Authors
2026-05-17
A quiet preprint server is now the front line of the AI backlash in science. ArXiv, a core infrastructure for physics, computer science and mathematics, is moving from vague warnings to explicit sanctions, announcing that authors who submit papers effectively written by large language models can be barred from posting for an entire year.
The platform’s stance is blunt: a chatbot cannot be an invisible coauthor. New guidelines stress that generative models may assist with grammar or translation, yet cannot be responsible for the design of experiments, formulation of proofs, or production of full sections of text, because such systems lack accountability, cannot sign copyright agreements, and may fabricate citations or claims. Policy language frames this not as a stylistic preference but as a matter of research integrity and traceable intellectual contribution, aligning arXiv’s rules with long‑standing authorship norms used by journals and funding bodies.
The real shock lies in the enforcement muscle. ArXiv signals that violations discovered through automated detection, cross‑checks with prior submissions, or community reports can trigger withdrawal of papers and a one‑year posting ban for all listed authors, not just the uploader. That collective penalty is designed to force research groups to build internal controls, from documented human review to lab‑level sign‑off on submissions, before any manuscript touches the upload form.
Loading...
