Axiomatic Architectural Safety for Superintelligent AI

Abstract

This paper is a technical companion to Benjamin Shapiro, The Decency Chip Manifesto — Protecting Humanity From Itself, first published on Zenodo on October 6, 2025. DOI: https://doi.org/10.5281/zenodo.17274356 This document was first published on the Thinking Software, Inc. website on July 25, 2026. Archived on Zenodo on July 27, 2026. Abstract: As artificial general intelligence (AGI) progresses toward artificial superintelligence (ASI), standard human-in-the-loop oversight mechanisms become technically unfeasible. Human cognitive bandwidth and execution velocity cannot operate at the speed of superintelligent runtime execution loops. Furthermore, historical precedents show that systems operating at higher levels of cognitive efficiency invariably alter, dominate, or displace lower-order intelligent systems. To prevent systemic existential risks without halting technological advancement, this paper proposes an architectural shift: Delegated Hierarchical Enforcement. Because human bandwidth cannot govern ASI directly, safety monitoring must be delegated to specialized, higher-order Guardian AIs operating under an immutable set of Axiomatic Core Invariants (the "Protocol Invariants").