Going by the screenshots, it looks like the censorship isn't that sophisticated.

Looks like it is just simple string matching, not even censoring based on the context of previous messages.