I've threat modeled this myself, and as I understand it the Bitwarden client side decrypts/encrypts everything locally. So even if backend was entirely compromised, it's never getting anything without the master password, and that's never sent across by the client. Then again, there's also the web interface.