Axiomatic Governance for Agentic AI: A Five-Layer Constitutional Framework Derived from First Principles and Validated in Production

Agentic AI systems—capable of autonomous planning, reasoning, and execution—are transi- tioning from prototypes to production deployment, yet existing governance frameworks provide prescriptive checklists without formal derivation: they specify what rules to follow but not why those rules are necessary or independent of one another. We present ORISMION, a five-layer axiomatic governance framework that derives all rules from a single objective function (survival maximization) through explicit inference chains grounded in thermodynamics, control theory, and epistemology. The framework identifies three irreducible architectural dimensions—cognitive (Ω), engineering (Σ), and governance (Γ)—established through six formal proofs, and enforces their recursive self-similarity at the code level: every API endpoint must exhibit a verifiable layered structure mapping directly to the axiomatic hierarchy. Validated in a production SaaS environment (309 automated tests, 28 architecture decision records, 18 FATAL-level rules), the framework’s extensibility was further tested through principled integration of insights from four frontier frameworks (GaaS, IMDA MGF, TRiSM, HAIG), expanding from 28 to 33 defense rules without requiring new architectural dimensions. The construction process itself—in which a non-technical founder and multiple AI agents collaboratively derived the formal framework— constitutes a documented case study in human-AI co-construction of governance institutions. We publicly release the theoretical framework and selected decision records; implementation-specific materials remain proprietary.