The risk of #2 (RMM breach --> client breach) is one of the main reasons that I was mostly looking at self-hosted RMM options when we were looking, and is the reason I wouldn't want to jump to a new-to-market RMM.
I still haven't hardened ours as much as I should (e.g. with the router completely blocking all non-US inbound traffic), but when things like the Struts vulnerabilities cropped up it was easy to tweak our router so that the only inbound connections allowed for the RMM were from the static or most recent IPs of our customers' offices. That cost us a few days of connectivity to anyone using laptops outside the offices or possibly from a dynamic IP, but that's only a few systems for a limited time.
For that matter I'd love to see some extra paranoia options - endpoints require pinned certs, ability to have the endpoints port knock if the initial connection fails, etc. I know port knocking isn't really security as such in that it's not encrypted, but if you have a service that accepts reporting from endpoints then by definition you have an open inbound port that can be probed. Port knocking means that it's a lot less likely that someone can start probing that port for issues (e.g. Heartbleed).