Clients can always be compromised. I'm not talking about a client that can't be compromised, but simply a client that is not compromised out-of-the-box.
That seems orthogonal to the subject of this discussion, i.e. "Compromise of the client side application or OS shouldn't break the security model."
That seems orthogonal to the subject of this discussion, i.e. "Compromise of the client side application or OS shouldn't break the security model."