Isolated Artificial Intelligence (IAI): A Privacy Preserving, Offline, Policy-Governed Architecture for Enterprise Knowledge Retrieval and Decision Support

This concept paper proposes Isolated Artificial Intelligence (IAI), a modular enterprise architecture for retrieval-augmented question answering and operational knowledge assistance inside a controlled on-premises or physically air-gapped environment. IAI combines heterogeneous data ingestion, local embeddings and retrieval, a local language model, identity-aware authorization, evidence-linked responses, and AI management controls. The research problem is not whether local language models exist, but how to integrate them into a repeatable, domain-agnostic system that can enforce departmental access boundaries, handle technical diagrams and structured operational records, and demonstrate measurable privacy, retrieval quality, reliability and resource efficiency. This publication defines a proposed design, requirements, threat model, experimental methodology and falsifiable evaluation criteria. Implementation, performance and security claims remain hypotheses pending independent testing.

Keywords: isolated artificial intelligence; offline AI; retrieval-augmented generation; local large language models; data sovereignty; enterprise knowledge management; access control; ISO/IEC 42001; design science research.