In this ART Talk, Dr. Romain-Daniel will explore two fronts of the reproducibility issue in animal research. The first dimension is statistical, where chronic underpowering, misuse of tests, pseudo-replication, and opaque reporting, lead to studies that can hardly be evaluated, let alone replicated. Compliance with reporting guidelines like ARRIVE remains insufficient despite years of advocacy.
The second front is newer. Generative AI is now integrated in research pipelines, from "vibe-coded" analyses that authors cannot interpret or verify, up to fabricated data, synthetic images, and industrial-scale outputs from paper mills. I argue that fixing one without the other is insufficient, and outline concrete lines of defense such as stronger (and down to earth) governance, mandatory sharing of data and code, and tools for detection, that may address both simultaneously.