November 1, 2025
LLMs flood, arXiv slams the gates
arXiv No Longer Accepts Computer Science Position or Review Papers Due to LLMs
ArXiv to CS writers: No more preprint thinkpieces—bring peer review
TLDR: ArXiv’s CS category will only accept review and opinion pieces that already passed peer review, blaming a surge of AI-generated surveys. Commenters are split between cheering quality control and mourning the preprint spirit, with confusion over whether this blocks early sharing and non-open journal papers.
ArXiv’s computer science category just put up a velvet rope: review and opinion pieces now need proof they’ve already passed journal or conference peer review before hitting arXiv. The site says it’s not a new policy, just stricter enforcement—blame the flood of AI-assisted “surveys” clogging the queue. Cue the community uproar. One commenter immediately fact-checked the headline, insisting the change is narrower than a total ban, while another cracked, “Isn’t arXiv where everyone uploads their paper?”—translation: the preprint party’s over, folks. The spiciest debate? Whether this nukes arXiv’s preprint vibe. Some argue it’s quality control against LLM (large language model) churn, accusing paper mills of pumping out annotated bibliographies with zero insight. Others, like Sharlin, claim it effectively shuts the door on preprints and even accepted papers unless the journal’s open, calling it a huge shift for early sharing. The mood swings from eye-rolls to existential dread: are we protecting science or gatekeeping it? There’s even a crypto-flavored pitch for a reputation system—publish with a public key, let trust follow the author—because CS will figure it out, right? For now, arXiv plays bouncer, asking opinion and review authors to bring receipts, while the crowd argues whether that’s saving science or killing the vibe.
Key Points
- •arXiv’s CS category will only consider review/survey and position papers that have been accepted at a peer‑reviewed journal or conference, with documentation required.
- •arXiv says this is not a policy change but a stricter enforcement of existing content types; such papers were never officially accepted content.
- •The shift responds to a large influx of review/position submissions, attributed in part to generative AI and large language models.
- •Many recent review submissions are described as annotated bibliographies lacking substantive discussion of open research issues.
- •The aim is to help readers find valuable expert work and free moderators to focus on arXiv’s core mission of sharing original research quickly.