Generative AI is already in production, newsroom, and streaming workflows, connecting to internal systems and beginning to execute tasks. What must be protected now extends beyond information to judgment, rights, and the operational permissions granted to AI. This session organizes AI adoption into four models based on who owns each layer of the stack, and shows how three core problems — the inability to explain, to stop, and to trust — appear differently in each model, and how the response changes accordingly. The scope of what can be delegated to AI will keep expanding. Accountability will not go away. We offer a map for confronting that question.