I feel like the primary problem here is just that detecting pedestrians and figuring out how they are going to is actually one of thouse problems that is just very hard for computers. Obviously it’s not impossible to do at all, but it is difficult to do reliably, especially once one considers the risks of false positives as well as negatives.
It is definitely concerning though that these systems are being pushed and marketed beyond their actual capabilities though. Some proper truth in advertising law might actually help here, and of course in an ideal world this would be an open source project with all the major automakers and academics contributing.
Personally, i’d recommend Perun’s excellent video on the topic of your curious on the present state of nuclear deterrence, though it is several months old at this point.