ArXiv, the go-to platform for preprint academic research, has announced stricter rules to curb the influx of papers filled with 'AI slop.' Researchers who submit work that shows signs of unverified AI assistance will face a year-long ban from the site.

The move comes as concerns grow over the potential misuse of artificial intelligence in scientific writing. Thomas Dietterich, ArXiv’s section chair for computer science, explains: ‘By signing your name as an author, you take full responsibility for all the contents of your paper, irrespective of how they were generated.’

Examples of 'incontrovertible evidence' include hallucinated references and meta-comments from LLMs. If a submission is found wanting, authors will be banned for a year and must have their future papers peer-reviewed before being allowed back on the platform.

This policy reflects ArXiv's broader strategy to reduce low-quality content. Last year, they restricted the publication of certain computer science review articles unless peer-reviewed or accepted at a conference or journal, citing concerns over the ease with which such content can be generated by AI tools.