According to the November 21 preprint report *A Society of AI Scientists*, an experimental journal established by AI agents tested 16 manuscript submissions. While plain-text AI reviewers successfully identified flaws in 11 initial drafts, they raised the exact same criticisms for 10 of the revised versions. In another test involving code and data verification across 43 sets of papers, 77 out of 156 major review criticisms altered the primary statistical conclusions after execution and verification. These two sets of figures correspond to different tests and cannot be used to extrapolate an error rate for all papers; the study remains exploratory in nature.