"Censorship", meaning: "malicious server or ISP silently blocks or withholds traffic from you" is a risk in Matrix today, completely independently of the reputation stuff being discussed here. The mitigation is to get rid of servers (and even ISPs), as per https://matrix.org/blog/2020/06/02/introducing-p-2-p-matrix/
"Censorship", meaning: "your server admin subscribed to a blocklist of child abuse content published by someone like iwf.org.uk" would stand out to users by the server publishing the names of the blocklists that their server admin has deployed. I'd call this server-side filtering or something instead, given the filters are visible. If the server admin withholds info about their filters, then you're back in the traditional sense of censorship from the prior paragraph.
We’ve been wargaming through all the various ways this could go horribly wrong (and have a few fun scenarii off the back of it - look for the GPT-3 example below), but on balance it feels a lot better than the Black Mirror episode where encryption is fatally weakened...