For any detection logic one has to take the video, .i.e. the series of images in, feed to logic of code, and return results. For a hidden or very big program, the code can as well be,

if nudity send_to_apple_server() print(“We detected nudity, and flagged this video”)

The user cannot differentiate it from well intended code. The right thing to do is not track at all! No “SMART” logic to “HELP”!