Damn, that's some impressive and crazy stuff.
With all the modelling software that is used for their automated detection you would have thought that it would also pick up on additional things like this. Surely they have a pattern of transactions over any period of time and could trigger an alarm when it doesn't match that pattern.
Of course if they have already compromised the internal systems which use the same credentials as the monitoring servers then they may, over time, change the triggers anyway which puts you back to square one in fighting them. Training the staff not to fall for phishing attacks. Obviously a lot harder in reality than it sounds. Most of the people we follow up with as their accounts have been compromised deny ever following the links we have them tagged in the firewall for.
Another method is good (and consistent) segregation of systems. Once again, data flow analysis in between each system on fibre taps with completely stand-a-lone systems doing the monitoring helps to increase the likelihood of keeping at least the monitoring solution secure, and so increases the ability to detect fraudulent non-analogous transactions. Yes it is expensive, Yes it is not necessary for most companies, we only tap our external lines, but then again, they are a bloody bank!