Researchers have reviewed over 200 publications on volumetric radiology AI, highlighting the limitations of current multimodal large language models (MLLMs) in preserving three-dimensional information. These models are often conditioned on two-dimensional images or report-derived text, which may not be sufficient for clinical interpretation. The authors propose a framework to assess the technical, workflow, and clinical credibility of volumetric radiology AI systems, emphasizing the need for faithful volumetric representation, traceable system behavior, and clearly defined human oversight in realistic workflows. --- Why it matters: This matters to researchers in AI because it highlights the limitations of current MLLMs and proposes a framework for evaluating the technical and clinical credibility of volumetric radiology AI systems. This has implications for the development of more accurate and reliable AI models that can integrate multimodal information and provide faithful volumetric representation. Source: https://arxiv.org/abs/2608.20549