The most obvious answer for me is that the computer could tell people to take interest if the email matches some criteria. So it isn't the intrusive one, but it could automate finding people to intrude on.
Do you think that this hasn't been happening? As someone posting on tech site, you should be aware that they scan emails for keywords, probably google and certainly the NSA.
Furthermore, training LSTM networks on emails has nothing to do with what you are concerned about.
Furthermore, training LSTM networks on emails has nothing to do with what you are concerned about.