You'd have to be pretty quick to get information out of the phone before the user catches on, especially if you've just asked the user to watch a video.

I just don't see this as an effective means of controlling a phone, when your primary means of "infection" involve user action.

The paper even acknowledges this, though almost dismissively: "One potential limitation of GVS-Attack is that the victim may notice the voice command played by VoicEmployer and interrupt it."

That's like, the whole reason this attack isn't very useful except in specific situations. No one's going to be running their botnet via GVS-Attack, that's for sure.