ArXiv, the dominant preprint server for physics, mathematics, and computer science research, will ban researchers who systematically submit papers filled with AI-generated content showing no human review. The platform targets papers containing "incontrovertible evidence that the authors did not check the results of LLM generation," including hallucinated citations, nonsensical references, and metadata comments accidentally left by language models.
The policy addresses a growing problem on ArXiv, where low-quality submissions using unchecked AI outputs have crowded the platform. Researchers report spending more time filtering garbage to find legitimate work. Hallucinated references stand out as obvious red flags. When papers cite studies that don't exist or attribute quotes to wrong authors, it signals authors never validated the LLM's output.
ArXiv's enforcement approach focuses on intent. Accidental AI-generated errors won't trigger bans. The platform distinguishes between researchers using AI as a tool while maintaining quality standards and those uploading wholesale AI garbage without verification. This nuance matters. Many legitimate researchers now use LLMs for drafting or editing while still reviewing content thoroughly.
The policy creates accountability without blocking AI use entirely. Researchers can employ language models in their workflow, but they cannot abdicate responsibility for accuracy. This sets a clear standard: use AI, but verify everything before publication.
The move reflects broader tension in academic publishing. ArXiv operates with minimal moderation compared to peer-reviewed journals, relying on researcher self-governance. As AI generation tools proliferate, maintaining quality requires stronger enforcement. Without intervention, the signal-to-noise ratio deteriorates, making the platform less useful for the research community.
ArXiv's action sends a message to researchers: automation is fine, but accountability is non-negotiable. Papers must reflect genuine intellectual work, not algorithmic output laundering. The policy preserves ArXiv's value as