A third-year medical student told me last month she was not going to apply to radiology. She had been interested all through second year, had done a sub-internship she liked, and had the grades to match anywhere. Then she changed her mind. The reason, almost verbatim: “My advisor said I should not go into a field that AI is going to replace.” She is not the only one. I have heard this from at least five students in the last year, from family medicine attendings warning their best students off radiology, and from internal medicine residents asking me, half-jokingly, whether I am worried about my career. The fear is real, widespread, and wrong. It is also doing damage to a field that is in shortage and that will be leading AI deployment to the rest of medicine for the next two decades. Here is the data the fear is built on. A recent analysis I co-authored of every AI/ML-enabled medical device the FDA has authorized over the past three decades, 1,430 devices, found that 76.5 percent were reviewed by the FDA’s Radiology panel. Cardiovascular came in second at 9.5 percent, neurology third at 4.5 percent. The remaining 19 review panels combined accounted for less than 10 percent. Pathology had nine. Microbiology had six. The psychiatry panel had zero. If you only saw the radiology number, the conclusion writes itself: AI is coming for radiology first and hardest. That is the version of the story that has reached advisors and students. It is also the wrong reading. That 76.5 percent represents where the data is, not where AI is replacing physicians. Radiology runs on DICOM, a universal imaging standard that makes a chest CT in Boston and a chest CT in Bakersfield the same file, labelable by the same workflow. There