At least for ML models whether encrypted or signed or grabbed from clearnet or through hacks become just another set of inputs to train on. If it helps classification it'd be used.

My intuition struggles with the many variables in play.

What motivated your initial note?