How are they supposed to filter out the fact that people "hide, cheat and lie to make money"?

This fact is an innate part of their training data.