Agentic Object-Oriented Programming (AOOP) and Asymmetric Agentic Governance (AAG): A Practical Architecture for System Sovereignty
Year: 2026
Type: preprint
Abstract: The deployment of Large Language Models (LLMs) in enterprise software ecosystems necessitates formal architectural constraints to separate probabilistic inference from deterministic state mutations. We identify Parametric Myopia (PM) as the systemic pathology - locally plausible but globally invalid refactoring - and Architectural Divergence (AD) as its measurable temporal manifestation. This paper proposes Asymmetric Agentic Governance (AAG) within the Agentic Object-Oriented Programming (AOOP) framework. We define a dual topology: a heuristic exploratory phase with read-only access for hypothesis generation, and a strictly bounded mutative phase subject to static and dynamic validation before any state transition. Under explicit Trusted Computing Base (TCB) integrity assumptions, the architecture enforces that outputs from the probabilistic domain cannot deterministically mutate system state without passing (1) static formal checking via AST Solver - YAML constraints compiled to Fragmented MSO to DUTA for O(N) validation via Courcelle's Theorem, and (2) dynamic runtime analysis via seccomp-bpf and eBPF instrumentation. State is defined as a left-fold over a cryptographically immutable event ledger with L2 sealing (HMAC-SHA256 in TEE/KMS) and 2-of-N multi-sig overrides. 44 pages, 7 formal specs. Licensed under CC BY-SA 4.0. Establishes System Sovereignty: verifiable, reproducible state execution.
Author: Terrazas A., J. Jesus
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7170716990
Cites:
Cited by:
Related to: