The_Fire
Unimpressed Female
I suspect that the selection criteria will be rather more sophisticated than simply looking for incidencences of (say) BOMB NUKE LONDON SECRET TOMORROW. One would anticipate an engine which looks at the incidence of associated words, cross-refers it back to previous e-mail from the same user, identifies any consistent patterns, and only then flags it up to a someone.
But I'm an architect. I really am only speculating!
As the part of me which is a web-programmer, I can see how that can work. At least partially. The part I'm having problems with is still the amount of information if we are talking about a group of people the size of eg. the complete number of residents in the US.
The pattern recognition of specific sentences could work, but in order to cross reference it back to previously sent and possibly deleted email, one would have to have one hell of a data storage warehouse. We are talking about more than terrabytes of old data here.
One would basically have to backup EVERYTHING ever sent or published on the internet to prevent deletion of possible relevant data. Not to mention the fact that you would need backup facilities for those data. And that's before adding possible recorded phone calls to the mix.
Again: To make it plausible (read: practical) one would need to narrow it down to persons of interest.