When agents delegate to other agents, can authority only decrease?

If your procurement agent can approve purchases up to $5,000, any sub-agent it calls should inherit that ceiling — or lower. Never higher. Privilege escalation through delegation chains must be structurally impossible, not just unlikely.

This is a direct consequence of agents creating intent rather than forwarding it. When Agent A delegates to Agent B, the authority transfer needs to be explicit, bounded, and auditable. Does Agent B know its ceiling? Does the infrastructure enforce it? Or does it just assume the delegation was reasonable?

The architecture question: can your system guarantee that authority monotonically decreases through any chain of delegation? Or are you relying on each agent to respect limits that were never structurally imposed?

Go deeper: Why Traditional IAM Breaks Down covers delegation chains, bounded authority, and why the infrastructure needs to enforce what policy cannot.