Decoupling Intelligence from Identity: The CPAP Protocol, the AI Ledger, and a Zero-Trust Middleware for Sovereign Enterprise LLMs

Enterprise adoption of Large Language Models (LLMs) is constrained by two structural vulnerabilities: data sovereignty risks when transmitting proprietary data to third-party cloud providers, and API drift caused by unannounced backend model updates that silently degrade static prompt systems.

We present the Model Continuity Architecture (MCA)—an operational framework developed by Chitrangana and deployed across enterprise clients between October 2025 and July 2026. The architecture decouples enterprise reasoning from cloud model identity through three integrated mechanisms:

(1) Context-Preserving Anonymization Pipeline (CPAP) executing deterministic Reversible Semantic Masking (RSM) to transform sensitive entities and magnitudes into pseudo-tokens while preserving mathematical topology;

(2) Dynamic Orchestration Router with cost-to-performance task pinning and automated circuit-breaking failover;

(3) AI Ledger—a localized, self-updating instruction accumulation system that converts operational failures into permanent deterministic constraints via human-in-the-loop auditing.

This technical report presents the complete architectural specification, formal design rationale, and reproducible synthetic benchmark evaluating core components. All performance claims reflect aggregated operational telemetry from active enterprise deployments. Due to binding client confidentiality agreements, specific organizational identities and proprietary data schemas are not disclosed.

The open-source benchmark suite and reference implementation are included in this upload.

A preliminary version of this work was previously published on SSRN (DOI: 10.2139/ssrn.7236658 ).