Too complicated
It seems to me to be pretty simple to do. Even if it isn't what Rudd thinks she wants.
Just give each post an AI scan between the time the poster hits SEND and the message appears. If it is "approved" then it just passes into the general population of pointless witterings.
However if it fails that first test, there are all sorts of possibilities. The bluntest instrument would simply be to lose the post (maybe Yodel would tender for that job?). But that might not be subtle enough. A far more amusing possibility would be to alter the post somehow.
I expect everyone has, at some point or other, typed "now" instead of "not" [ Dilbert reference goes here ]. And I am equally sure there are far more devious possibilities that could be laid at the door of predictive text.
So, it would not be beyond the bounds of possibility to take messages that would and could be flagged as "terrorism" and turn them to your own advantage. Thus: I do now believe America is the source of all that is good from someone who's ovine followers would then be forced to give up the fight and spend, spend, spend.
With luck the originator wouldn't even notice the change!