Earlier quoted context omitted.
I'd guess that the spam filters probably go crazy because users actually mark this kind of email as spam. If so, this is the system working. People say something is spam, and less of that kind of email gets sent.
The most likely reason the spam filters go crazy because there is a huge amount of bona-fide ICO and blockchain related spam, and they can't tell the difference between that and emails that users actually want.
But the way we detect bona-fine spam is by user reports. We then train models / decide heuristics based on these reports, but in the end, users decide what they report.