The type of monitoring used was disclosed as was the "likely cause" of me being flagged. This wasn't a secretive kind of thing. Of course one can't "know" anything as it's a black-box. But it's not a stretch to see how such systems may conclude, from automated profiling, that users of privacy-centric services are more likely to be associated with fraud and hence flagged. This also represents a more general issue with the application of machine learning.