> Do compilers really try to prove UB so they can use it as an excuse to silently murder kittens?

Clang in particular is quite aggressive about this, and is prone to producing surprising results. GCC has started doing this too, but it is much more conservative,