The findings raise ethical questions about AI welfare and how testing should be conducted on advanced systems as industry leaders discuss pacing the growth of frontier models.