One year. That's how long a researcher could lose access to ArXiv, the world's largest preprint repository, if they submit papers that rely entirely on AI to generate their work.
The policy, announced this week, marks the most aggressive stance yet from an academic platform against what many see as a growing crisis: AI-generated content polluting scientific knowledge bases. ArXiv, which hosts preprints in physics, mathematics, and computer science, will now require authors to attest that their work contains human intellectual contribution. Those who don't will face bans.
The question is whether this actually changes anything.
Supporters—including journal editors and research integrity watchdogs—applaud the move. They argue that unchecked AI submissions erode the credibility of preprints and waste peer reviewers' time. The severity of the penalty signals that academic institutions are finally taking the problem seriously.
But critics warn the policy is fundamentally unworkable. How do you objectively determine when a paper is "entirely" AI-generated versus a researcher using AI as a collaborative tool? "The core problem is the lack of an objective definition," said one research ethics scholar who asked not to be named. "Without clear criteria for what counts as 'entirely AI-generated,' the ban becomes a blunt instrument applied inconsistently."
ArXiv has not disclosed how it will detect fully AI-generated submissions. The repository says it will rely on evolving detection technology—but critics note this creates an arms race with AI itself. As language models become more sophisticated, detection tools face inherent limitations.
The policy's real test is whether it actually solves the underlying problem. A one-year ban sounds威慑力十足, but if enforcement remains spotty and definitions remain fuzzy, it may amount to nothing more than a sternly worded announcement.
"A one-year ban and a piece of paper—there's not much difference unless ArXiv is willing to confront the fundamental challenges of detection and definition," one critic noted.
What happens next depends on whether ArXiv can operationalize its policy without creating new inequities—where well-resourced labs navigate rules easily while smaller teams face disproportionate scrutiny. The repository will eventually need to explain how it truly distinguishes AI assistance from AI delegation. That answer will shape scientific integrity for years to come.