arXiv, the largest host of pre-peer-reviewed scientific papers, will now ban scientists for a year if their research contains hallucinated references or other obviously AI-generated content.

The research repository’s chairman said authors have “full responsibility” for the content of works published in their name.

AI is boosting scientific research, but also facilitating fraud: Submissions to one journal have risen 42% since ChatGPT’s launch, and the writing has become worse, Forbes reported, with many papers detectably AI-generated.

Computer science, the most AI-exposed field, is worst affected.

arXiv has already banned all comp-sci review papers over an “unmanageable influx” of fakes, and one researcher warned that it took just 54 seconds to create a fake paper about an experiment he never performed.