This in turn implies to me that anything server side needs to be more than just enforcement i.e. performing an authorization check at the time of an attempt to access the model / execute an operation is the bare minimum - we potentially need to be able to go beyond this to pro-actively identify what can or can not be accessed.
Enforcement might be a misleading term, but this is what I meant, when talkig about the CLI and the Console. Somehow clients need to bootstrap themselves from the model in order to suppress interaction elements.