In carry-on bags - bags go through an x-ray machine with an operator looking at so many negatives that it's next to impossible for them to spot a positive case.

Schneier has a nice piece about it in an excerpt from one of his books: https://www.schneier.com/blog/archives/2005/04/failures_of_a...

Makes no sense why ML isn't used, with a quick peep at a potential positive by a human before they rip your luggage open with a buck knife.

>Makes no sense why ML isn't used

I agree, although remember that it changes the problem for the attacker too: they're switching between trying to fool a (bleary-eyed) general intelligence to simply fooling a ML algorithm... think about how obvious and jarring anti-facial-detection face camoflage is for a human, whereas the machine simply sees nothing.