Read paper online Download PDF
Enterprise adoption of generative AI has outpaced the knowledge infrastructure needed to operate AI systems reliably. A team can build an impressive assistant over a curated folder, a small policy corpus, or a narrow set of service tickets. The harder work begins when the same assistant is expected to operate across document repositories, SaaS platforms, collaboration tools, business applications, and data estates with uneven governance. In that setting, many failures that look like model failures are actually failures in the knowledge path around the model.
This article introduces GKS-5, a five-layer reference architecture for Governed Knowledge Systems in enterprise AI. GKS-5 separates five responsibilities: data ingestion, storage and semantic indexing, retrieval, model orchestration, and application integration. It then pairs those layers with six cross-layer controls: metadata, authorization, freshness, observability, evaluation, and versioning. The purpose is to reduce knowledge integrity drift, defined here as the gradual loss of alignment between source systems, metadata, permissions, indexes, retrieval behavior, orchestration logic, and the outputs users see.
The contribution is a practical architecture and evaluation framework for enterprise AI knowledge systems. It includes a failure-mode taxonomy, a readiness scoring model, an evaluation protocol, and a scenario-based validation using the common case of an internal policy assistant moving from pilot to production. The core claim is deliberately simple: enterprise assistants are rarely durable products by themselves. They are interaction surfaces over deeper knowledge platforms. Better models help, but they do not remove the need for governed ingestion, authorization-aware retrieval, source authority, traceable context assembly, and systematic evaluation.
This technical report introduces GKS-5, a reference architecture for governed enterprise AI knowledge systems. The paper argues that durable enterprise AI requires more than model selection or prompt engineering. Production systems need reliable ingestion, semantic and lexical indexing, authorization-aware retrieval, orchestration controls, metadata discipline, freshness tracking, observability, evaluation, and versioning.
The paper is intended for AI platform leaders, enterprise architects, data and AI governance teams, RAG platform teams, and organizations moving generative AI systems from pilot demonstrations to governed production platforms.
Enterprise AI; Governed Knowledge Systems; GKS-5; Retrieval-Augmented Generation; Knowledge Architecture; Semantic Retrieval; AI Governance; Agentic Systems; Knowledge Integrity Drift; AI Platform Architecture.
You can read the paper below or download the PDF.
Sure, R. W. (2026). Enterprise AI Is a Platform Problem: GKS-5, A Reference Architecture for Governed Knowledge Systems. Independent technical report.
PDF: GKS-5_arXiv_v0.5.pdf
Related work: AGL-1: The Enterprise AI Governance Layer as a Control Plane for Trusted Enterprise Intelligence
Related work: CAGE-1: Control, Assurance, and Governance Evaluation for Enterprise Agentic AI
See the PDF for the full reference list.