arXiv says it will stop accepting computer science papers that haven't been vetted by an academic journal or a conference, after a surge in AI-generated papers

Cornell University's academic paper repository will no longer accept Computer Science papers still under review.

404 Media 2025-11-03 Matthew Gault

Context & Ripple Effects

arXiv’s role as a rapid-sharing venue puts its Computer Science intake rules at the center of a quality-control problem: AI-generated submissions can reach readers before conventional review. Earlier criticism of undisclosed AI-generated conference submissions showed that the pressure extends beyond repositories.

The move is an early gating response in a coverage arc that later includes one-year bans for papers with incontrovertible AI-generated work. It matters because arXiv is tightening access to a major distribution channel rather than merely asking authors for disclosure.

First-order effects

Computer Science authors whose work remains under review lose arXiv as an immediate preprint route and must first clear journal or conference vetting.
arXiv shifts more of its CS screening burden to established academic publication venues, reducing the pool of unvetted submissions it accepts.

Second-order effects

Journals and conferences become more consequential gatekeepers for research seeking arXiv distribution, while authors may face slower public dissemination in CS.
The rule creates incentives for other research repositories and preprint categories to clarify their own AI-submission controls; later arXiv coverage suggests enforcement can progress from intake restrictions to author penalties.

Third-order effects

If similar controls spread, open preprint infrastructure may evolve from an access-first commons toward a more actively governed research corpus with provenance and quality checks embedded in distribution.
The trade-off will be durable: stronger filters can protect trust in research discovery, but they can also make legacy review systems a prerequisite for visibility and narrow the speed advantage of preprints.

The trend: AI-generated research volume is pushing scientific publishing infrastructure to replace broad open submission with operational controls that protect the credibility of shared knowledge repositories.

Discussion

@stephenlcasper Cas on x
🧵 It's unfortunate that AI slop is polluting scientific lit so much. I'm glad @arxiv is trying to address this. But I think the cure may be worse than the disease. 🧵Here is a thread of 7 reasons I disagree.🧵 https://blog.arxiv.org/...
@stephenlcasper Cas on x
I get that @arxiv is a free public good run by unpaid and unthanked volunteers. I appreciate it. But I think it might be in over its head. I'd gladly pay a fee (e.g. $20) per submission if it allowed for reasonable moderation...
@tdietterich Thomas G. Dietterich on x
Again, we lack the moderator resources to separate the good from the bad, so we will be relying on peer review to do this prior to consideration by arXiv.
@tdietterich Thomas G. Dietterich on x
We have also faced a big increase in position papers (due primarily to the new Position Paper tracks at AAAI, ICML, and NeurIPS). Official policy has been that opinion papers are not appropriate for arXiv, but some of these are of very high quality and research relevance.
@tdietterich Thomas G. Dietterich on x
We were driven to this decision by a big increase in LLM-assisted survey papers. We don't have the moderator resources to examine these submissions and identify the good surveys from the bad ones.
@danrosen.xyz Dan Rosen on bluesky
interesting how this isn't a new rule, just the enforcement of existing rules. I tend to think pre-print culture is a bit cavalier and this is why [embedded post]

Chronicles