arXiv Bans Hallucinated References: One Year Suspension for Fake Citations
If you thought AI could write your references for you, get ready for a year-long timeout.

We've all been there: you're reading a paper on arXiv, reach the references section, and half the links lead nowhere. Or worse, they point to completely made-up works. It's like the authors asked GPT to fill in the bibliography while they went for a coffee break. Well, arXiv has had enough — from now on, hallucinated references can get you banned for a whole year.
The new policy, hinted at by Tom Dietterich on Twitter, is a harsh but necessary move. If a paper is found to contain references that don't exist, the author gets a 12-month suspension from arXiv. Imagine not being able to post your preprint while competitors are already sharing their ideas. Tough, but perhaps fair.
Of course, there's a catch: how will arXiv detect fake references? Manual review of every paper would require an army of editors. Most likely, they'll implement automated validation of DOIs or URLs. But what about references to unpublished works? That might require a trust system or a disclaimer.
For developers and researchers, this is a clear signal: stop relying on LLMs for critical tasks like citations. A neural network can generate a convincing list of references, but that doesn't mean they're real. Better to double-check yourself than to face a year-long ban.
METABYTE studio's take: We're not arXiv, but we value honesty in code and documentation too. If you need help automating link validation or building tools for scientific integrity — we can not only hallucinate new features but actually implement them for real.
NEXT STEP
Liked the approach?
We apply the same principles to client projects: AI, automation, products that don't die after launch.