Final anecdote. A certain attack requires the fraudster to call the target's bank and ask a few questions. For some reason in ALL of these calls there is the sound of a baby crying in the background.
Our theory is that while testing the attack vector, the first time it 'worked' a baby must have been crying. The fraudsters think it works as some kind of high frequency disruption to confuse any biometric systems that are processing the call, so they play a RECORDING of a baby in the background of all subsequent calls.
It reminds me of learned behaviour in animals. The pigeon stands on one leg and gets a treat. The pigeon now thinks the one legged approach is what makes the treat appear.
The icing on the cake was when I got a call from a bank asking if there are any biometric systems that can detect the sound of crying babies...