Right now, even the FBI can remotely activate the microphone on your phone. Doubt this will change anything in the short run.
The NSA never cared about listening to everything, that would be completely unmanageable, flagging conversations with key phrases has always been the course of action.
>Not to mention trivial to detect. Just watch CPU consumption and start spouting off trigger words.
It'd also be trivial to detect that you were using a CPU usage monitor and turn off those keywords, viruses do these sorts of things all the time to avoid detection.
Also, if you want an undectable CPU monitor, just use a temperature sensor strapped to the processor. Temp-monitoring was even used to find TOR hidden services.
For any kind of system to be workable, it's likely they'd have to narrow it down considerably to suspect users. Someone calling the phone number of a person or organization known to be N-degrees separated from real known targets is more likely to yield fruit than scooping up voice conversations from every phone that heard the phrase "bomb making"
Operating a roving bug does use the phone's battery when active, of course. The tricky part is that can be activated when the phone is nominally off, probably by means of the phone waking up periodically and listening for special commands from the network.
If my surmise is correct and this feature is built into the baseband, depending on the system architecture and power management, a roving bug might use only a very small part of a smartphone's relatively large battery.
Roving bugs require secret cooperation from network operators and from chip makers (baseband firmware engineers, specifically). Makes one wonder how well these confidential cooperative agreements have held up as more chip makers are located in the PRC.
http://www.youtube.com/watch?feature=player_embedded&v=bUmuI...