with enough samples, it shouldn’t be hard to extract a content filter’s opinions, for example you could use your botnet to map what twitter really considers bannable and then generate maximally bannable tweets