Self-declaration systems for AI agents have a fundamental problem: they work best on the agents that need them least.