The liability arrangement in AI-assisted radiology is not an accidental gap — it is the predictable outcome of deploying tools whose vendors have structured away their own legal exposure. has not evolved a theory of shared responsibility; courts default to the human who approved the output. That default made sense before AI was doing substantive diagnostic work. It does not make sense when a radiologist is reviewing dozens of AI-flagged scans per shift, under time pressure, with tools whose reasoning they cannot fully interrogate.