ArXiv Now Requires Peer Review for Certain Papers as AI-Generated Submissions Surge

ArXiv is a massive online library where scientists share research findings. It just changed how it accepts papers in computer science. Starting in October 2025, if you want to submit certain types of papers — review articles that summarize existing research, or position papers that argue for a particular approach — you now have to show that other experts have already reviewed and approved your work at a journal or conference first.

This is a big shift. For decades, ArXiv let researchers share preliminary work, get feedback, and claim priority before formal publication. Now, for some paper types, that early-stage sharing is restricted. The reason: artificial intelligence is generating research papers at scale, and many of them are flooding the system.

What Changed and Why

The new rule affects only two specific kinds of computer science papers: review articles (comprehensive summaries of what's known in a field) and position papers (arguments for why the field should take a certain direction). Everything else stays the same for now.

ArXiv explained the change this way: it saw an "onslaught" of AI-generated papers, especially in computer science. These papers often read like legitimate research because modern AI is sophisticated enough to write grammatically correct academic prose. But they lack real insight or novelty. The platform had to act.

The language models that power tools like ChatGPT are so good at mimicking academic writing that old-fashioned spam filters do not catch them. This is different from the automated spam that plagued early internet forums, where keyword filtering worked. This AI-generated content sounds real.

The new rule applies only to computer science for now. ArXiv said it will watch other fields and may tighten things there too if they face similar problems.

A New Practical Hurdle

ArXiv used to be a quick way to get your research out there. You would draft a paper, upload it, and people would read and cite it within days. The platform thrived on speed.

Adding a peer review requirement first slows everything down. Journals can take months to review a paper. If you have to wait for that review before ArXiv accepts a review article or position paper, you have lost the chance to get early feedback or signal that you had the idea first. Researchers who rely on ArXiv to show they are working on a topic — especially younger researchers building a career — now face a real inconvenience.

This change may make synthesis work and broad methodological discussions circulate more slowly at a time when artificial intelligence itself is moving very fast. Some researchers will have to find other places to publish these kinds of papers, or change their strategies altogether.

What Else is Changing

Starting February 11, 2026, ArXiv will also require all submissions across all categories and languages to include an English version. This makes the platform more searchable and citable globally, but it adds another task for researchers who do not speak English as a first language.

The Bigger Picture

Academic and research institutions are grappling with AI-generated content across the board. Cornell University, which runs ArXiv, published reports in 2023 and 2024 about how to handle artificial intelligence in research and administration. Those reports likely shaped ArXiv's response.

The distinction between using AI as a tool — for editing, translation, or finding papers — and using it to generate the core research itself matters for how platforms enforce rules. Most researchers now use AI to help, and that is not necessarily a problem. Wholesale AI generation of scientific content, though, threatens the integrity of the scientific record.

ArXiv's approach is cautious. Rather than banning AI-assisted research outright, the platform targets the specific submission categories most vulnerable to automation. In some ways, this is a pragmatic fix to a real problem.

What Comes Next

Researchers in artificial intelligence, machine learning, and natural language processing will feel this change most directly. These fields have relied on ArXiv to share new techniques and broad surveys quickly — faster than traditional journals allow. Peer review requirements may slow that circulation.

The longer-term outcome is worth considering. If ArXiv successfully filters out low-quality, AI-generated papers, the remaining work on the platform might gain credibility. Researchers could trust what they find there more. That payoff could offset the friction of slower publication.

Still, language models keep improving. ArXiv and other academic platforms will likely face new challenges as AI capabilities advance. Maintaining research quality while preserving the speed and openness that make shared science possible is a tension that will not go away soon.

Language Policy Update

ArXiv will require English translations of all submissions starting February 11, 2026, regardless of the original language. This broadens access and citation practices but creates additional work for non-English speaking researchers.