I think the idea of "screening" is that you don't just race off to a mastectomy the minute some AI model goes off. Of course, putting more false positives through a fallible process of review does run the risk you speak of.