arXiv introduces a one-strike policy for AI-generated slop, imposing a one-year ban on authors who fail to verify LLM-produced content.
arXiv Tightens Rules on AI-Generated Submissions
arXiv, the premier preprint repository for scientific papers, has updated its Code of Conduct to clarify author responsibilities regarding the use of generative artificial intelligence (AI) tools.
The policy, announced by arXiv's computer science section chair, Thomas Dietterich, introduces a clear penalty for submissions containing evidence that authors did not verify content produced by large language models (LLMs).
Policy Details
The updated policy states that each author is fully responsible for all content in their submission, regardless of the method of generation. While the policy does not prohibit the use of LLMs, it establishes a critical boundary: if a submission contains incontrovertible evidence that the authors did not check the results of LLM generation, the paper may be deemed untrustworthy.
Enforcement and Penalties
Violations result in a one-year ban from arXiv, enforced under a strict "one-strike" rule.
- Process: Moderators flag the issue, section chairs confirm the evidence, and authors have the right to appeal.
- Reinstatement: After the ban period concludes, subsequent submissions must first be accepted at a reputable peer-reviewed venue before they can be posted on arXiv.
Examples of Evidence of Negligence
The policy provides concrete examples of evidence that would indicate a failure to verify content:
- Hallucinated references (non-existent citations).
- Meta-comments generated by the LLM, such as: "here is a 200 word summary; would you like me to make any changes?" or "the data in this table is illustrative, fill it in with the real numbers from your experiments."
Context
arXiv has previously taken steps to address the influx of low-quality AI-generated papers, including requiring endorsements for first-time posters. The organization is also transitioning to an independent nonprofit structure to raise funds to address issues related to AI-generated content.