A recent study published in BMC Medicine by researchers at Queensland University of Technology and the Australian Center for Health Services Innovation has raised significant concerns regarding the reliability of artificial intelligence models used for predicting stroke and diabetes risk. The investigation, led by Dr. Alexander Gibson alongside Professors Adrian Barnett and Nicole White, examined two widely downloaded health datasets hosted on the machine-learning platform Kaggle. Despite being marketed as a premier resource for AI development, the datasets lacked verifiable origin stories, collection methodologies, and patient verification, scoring zero out of nine on the internationally recognized TRIPOD+AI reporting framework for data provenance. The research team discovered that the datasets had been utilized in 125 peer-reviewed studies, with three derived prediction models already integrated into clinical workflows. One model was even cited in a medical device patent and referenced across 86 academic review articles. Notably, seven publications relying on these datasets have already been retracted due to unreliable findings. The authors described the data as exhibiting unusual statistical patterns that strongly indicate compromised authenticity and unsuitability for medical research. Dr. Gibson emphasized that prediction models constructed from unverifiable data pose a direct threat to clinical decision-making. He warned that without foundational data transparency, AI outputs remain unreliable and risk misleading healthcare professionals while potentially endangering patients. The findings underscore a systemic vulnerability in the rapidly expanding field of digital health, where high-volume research often prioritizes speed over basic scientific rigor. In response to the findings, the research team has called for immediate regulatory and editorial action. They recommended that academic journals, research funders, and data repositories enforce stricter disclosure requirements regarding data sources before accepting AI-related studies. Furthermore, the authors advised that the problematic Kaggle datasets be permanently removed to halt further propagation. The study has already prompted the inclusion of these cases in the Collection of Open Science Integrity Guides. As artificial intelligence tools become increasingly embedded in healthcare infrastructure, experts stress that robust provenance standards and enhanced oversight are critical to ensuring that AI-driven medical models remain safe, accurate, and trustworthy.