We live in an age where the voice can be processed locally on the phone (we’ve had on-device speech-to-text since the late 90s…), and it’s already listening for a wake word, meaning mic is always hot. It doesn’t need to be streamed and use bandwidth; it can fire off 4K of JSON every few hours and relay more than enough information.
Just program whole dictionary of key phrases and scan the wake word buffer like you are already doing. Easy, stealthy, encrypted. Every voice assistant from a major tech company could (and likely IS) doing this.
This also provides ample opportunity for domestic (or even foreign!) spying my state actors, too.
I don’t think they do it to make us dumber (there are other methods for that) but to make it accessible to people who don’t already know better.
So, in this case, inadvertent is correct; Although it’s not without side effects.