arXiv Tightens Computer Science Submissions: Review and Position Papers Now Restricted
arXiv will no longer accept most computer science review articles and position papers. The move is a response to a surge of low-effort, AI-assisted submissions that read like annotated bibliographies and add little to ongoing research.
The service remains a core preprint venue for rapid dissemination in fields like AI and machine learning. This change is about enforcing existing standards, not shutting the door on legitimate science.
What changed
arXiv is applying a blanket restriction in the computer science category: review/survey articles and position papers will be rejected unless they include documentation of successful peer review. Submissions without proof of prior peer review will likely be denied.
- Original research contributions remain welcome.
- Review and position papers must show evidence of peer review at a recognized venue to be considered.
- Moderators aim to refocus time on substantive work rather than AI-generated overviews.
Why it matters
In fast-moving areas like AI, arXiv is often the first stop for sharing results. The flood of LLM-written summaries has made it harder for readers and moderators to find work with real signal.
For labs and authors, this raises the bar on non-original CS submissions. If your manuscript synthesizes prior work or argues a stance, plan for journal or workshop peer review first, then post the accepted version with documentation.
If you're planning a CS preprint on arXiv
- Submit original research with clear, testable contributions, data, and code where possible.
- For reviews or position pieces, target a peer-reviewed venue first; include acceptance or successful review documentation with your arXiv submission.
- Be explicit about any AI assistance and focus on analysis, synthesis, and open problems-not just citations.
- Use arXiv to increase access and speed, but avoid treating it as a shortcut around peer review for non-original work.
Will other fields follow?
arXiv notes computer science is the current hot spot for low-value, LLM-produced overviews. If other categories see similar patterns, they may adjust moderation to protect readers' time and uphold scholarly value.
The bigger issue: AI-generated research noise
Pay-to-publish schemes and low-quality outlets existed before AI, but LLMs have amplified the volume. Even reputable processes have cracked: one journal retracted a paper that included an AI-generated image of a giant rat penis, and some peer reviewers have been caught leaning on chatbots under workload pressure.
The result is more noise for editors, reviewers, and readers to sift through. arXiv's policy enforcement is one attempt to push the system back toward substance.
Where to read more
For submission guidance and policies, start at arXiv's official site: arXiv.org. If you mentor teams on responsible AI use in research writing and review, you may also find curated training resources helpful: Latest AI courses.
Update
Earlier coverage suggested arXiv had stopped accepting computer science articles under peer review. That was incorrect. The restriction is narrow and applies to review/survey articles and position papers without documentation of successful peer review.
Your membership also unlocks:
