Faceted Authority Attenuation: A Product Lattice Model for AI Agent Governance

As AI agents gain the ability to take real-world actions under delegated human authority, governance systems must enforce that delegated capabilities can only be attenuated, never amplified. Current implementations enforce this principle through ad hoc, per-dimension checks — verifying scope inclusion, spend limits, and delegation depth independently. We show that these heterogeneous constraints are instances of a single structural principle: agent authority is an element of a product lattice, and delegation is a monotone function on that lattice. We formalize this as faceted authority attenuation over seven constraint dimensions (scope, spend, depth, time, reputation, values, reversibility), prove that the monotonic narrowing invariant holds over the product ordering, and demonstrate that this formalization yields three concrete benefits unavailable to scalar or ad hoc approaches: (1) structured, machine-readable denial diagnostics that identify exactly which constraint dimensions failed and why, (2) cryptographically signed authorization witnesses that capture an agent's position in the authority lattice at execution time, and (3) near-miss detection that warns when an agent approaches constraint boundaries before violation occurs. The formalization is implemented in the Agent Passport System (first published to npm February 2026), an open-source protocol with 1,634 tests across 85 modules, validated through cross-protocol composition tests with five independent working group members on production infrastructure.