arXiv Cracks Down on Careless AI-Generated Papers

Researchers who upload preprints to arXiv are facing stricter moderation rules aimed at sloppy or unverified AI‑generated content. The repository has clarified that submissions containing clear signs of unchecked large language model (LLM) output can trigger a one‑year ban for all listed authors.

The policy does not prohibit AI assistance outright. Instead, it reinforces a long‑standing principle of academic publishing: authors are fully responsible for everything in their paper, regardless of how the content was produced.

What Triggered the New Enforcement

As generative AI tools became widely used in research writing, moderators began seeing papers with obvious artifacts from language models—such as fabricated references or comments clearly intended for a chatbot rather than readers. These errors undermine trust in preprints because reviewers and readers cannot easily verify whether other parts of the paper are reliable.

To address this, arXiv moderators clarified how their Code of Conduct applies to AI‑generated material.

What Counts as “Incontrovertible Evidence” of Unchecked LLM Use

The policy focuses on unmistakable signals that AI output was included without verification. Examples cited by moderators include:

• Hallucinated or non‑existent references in the bibliography
• Leftover chatbot conversation artifacts, such as prompts or responses embedded in the text
• Meta‑comments from the model, for example instructions like “Here is a 200‑word summary” or placeholders such as “fill in the real numbers from your experiment.”

Such artifacts strongly suggest the authors did not review the generated content before submission. When moderators encounter this kind of evidence, they may conclude that the reliability of the entire paper is questionable.

Penalties Authors May Face

If moderators determine that a submission contains clear evidence of unchecked LLM output, several consequences can follow:

• The paper is rejected.
• All listed authors may receive a one‑year ban from submitting to arXiv.
• After the suspension, future submissions must first be accepted by a reputable peer‑reviewed venue before they can appear on arXiv again.

The policy applies to every author on the paper, reflecting arXiv’s rule that authors share responsibility for the content they sign.

AI Tools Are Not Banned

Importantly, arXiv has not banned the use of generative AI in research writing. The platform’s stance is that AI tools can be used—but authors must carefully verify and take responsibility for the output.

This mirrors broader academic norms: software may assist with drafting or analysis, but the authors remain accountable for accuracy, citations, and claims.

Review and Appeal Process

Moderation decisions on arXiv are not final without recourse. Authors who believe a decision was incorrect can appeal through arXiv’s moderation appeals process, asking moderators to reconsider the submission or classification.

In some situations, arXiv may require that a paper be accepted by a conventional journal before considering an appeal related to a declined submission.

Why the Policy Matters

arXiv is one of the most important infrastructure platforms for early scientific communication. The new enforcement approach reflects a growing concern across academia: generative AI can accelerate writing, but it also increases the risk of fabricated citations and subtle factual errors.

Rather than banning AI outright, arXiv’s approach sends a clear signal: AI assistance is acceptable, but unchecked AI output is not. The responsibility still rests squarely with the researchers whose names appear on the paper.