New arXiv policy: 1-year ban for hallucinated references
In a significant step to combat the rise of AI-generated fake citations, arXiv has implemented a policy that will ban authors for one year if their submissions contain hallucinated references. The policy, announced on Twitter by Tom Dietterich, a prominent figure in the AI community, targets the growing problem of papers that cite non-existent works, often produced by language models. This measure is intended to deter researchers from submitting manuscripts with fabricated references, whether intentional or due to AI misuse. The ban applies to the submitting author and may extend to co-authors, depending on the severity. arXiv, a preprint repository widely used in computer science and related fields, has seen an increase in such incidents as AI tools become more prevalent in academic writing. The policy is effective immediately, with enforcement through manual and automated checks. This move underscores arXiv's commitment to scholarly integrity and sets a precedent for other repositories.
Developers and researchers must ensure their citations are real or face a year-long ban from arXiv.