Software that already is listening has more vectors of attack to put it simply. I'm not a fan of features that compromise privacy and security for the sake of convenience.
That's really nonsensical. Nothing about using an audio input makes code more vulnerable. The interface is always there if the hardware is.