A microphone and some basic ML? Sort of tongue in cheek, since I don't know the range of state of the art acoustic side-channel taps. I guess you'd also probab

A microphone and some basic ML? Sort of tongue in cheek, since I don't know the range of state of the art acoustic side-channel taps. I guess you'd also probably have power fluctuations and network timing channels to exploit. Plus, a lot of different points at which to attempt to insert a second stage into the connected devices themselves, using all the tricks everyone else in this thread has mentioned.