Do they often publish transparency reports about their ML systems that cannot identify months-long spam campaigns where scammers post identical messages tens of thousands of times?