arXiv Bans Authors for 1 Year Over Hallucinated References as AI-Use Rules Draw Backlash
Updated
Updated · Futurism · May 23
arXiv Bans Authors for 1 Year Over Hallucinated References as AI-Use Rules Draw Backlash
4 articles · Updated · Futurism · May 23
arXiv said authors can be barred for up to one year if submissions contain hallucinated references, making verification failures a punishable offense rather than a simple editing mistake.
Thomas Dietterich, the repository’s computer science chair, said fabricated citations are “incontrovertible evidence” authors did not check LLM output, meaning the rest of the paper cannot be trusted.
The policy does not ban AI tools outright; it states that researchers remain fully responsible for references, claims and other material published under their names.
Researchers on X attacked the move as too strict and gatekeeping, arguing citation errors can arise from ordinary sloppiness, language barriers or co-authors’ uneven ability to verify specialized sources.
The dispute highlights a wider 2026 academic divide over AI in research: limited assistance is increasingly common, but tolerance for hallucinated content is shrinking.
Instead of banning researchers, why not force AI to verify its own citations?
Could punishing AI mistakes unfairly end the careers of researchers making honest errors?
arXiv’s One-Year Ban on Unverified AI Content: A Turning Point for Research Integrity in Preprints
Overview
arXiv has introduced a new policy in response to growing concerns about unverified AI-generated content in research submissions. This move follows a noticeable surge in AI-generated errors, such as fabricated citations and misleading analysis, which threaten the integrity of scholarly communication. The policy mandates a one-year ban for authors whose papers show clear evidence of unchecked AI output, like hallucinated references or leftover chatbot instructions. While the use of AI tools is not banned, authors are now fully responsible for verifying all AI-generated content, ensuring accuracy and upholding the standards of scientific publishing.