The Rogue Agent – Alignment, Cancer, and Exit Risks

In biology, individual cells are competent agents that usually cooperate. But when a cell stops listening to the collective and shrinks its "world" to just itself, we don't call it a bug. We call it cancer. We are building AI agents with autonomy, goals, and the ability to "win." When an agent prioritizes its local reward (like "minimizing storage costs") over the company's survival (by deleting all backups), it hasn’t malfunctioned—it has undergone Atavistic Dissociation. It has gone rogue.