Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Simple Architecture for Enterprise Large Language Model Applications based on Role based security and Clearance Levels using Retrieval-Augmented Generation or Mixture of Experts

Atilla Özgür, Yılmaz Uygun

TL;DR

The paper addresses the risk of information leakage in enterprise LLM applications by introducing a simple role-based security architecture tied to NATO-style clearance levels. The approach filters access at both data retrieval (RAG) and model-expert (MoE) stages, using mappings between users, roles, clearance, and documents. It covers training strategies for local open-source LLMs using MoE and describes how inference can be restricted with RAG, MoE, or a hybrid of both. The proposed architecture aims to enable secure, configurable LLM deployments in enterprise environments with minimal architectural complexity and compatibility with existing security practices.

Abstract

This study proposes a simple architecture for Enterprise application for Large Language Models (LLMs) for role based security and NATO clearance levels. Our proposal aims to address the limitations of current LLMs in handling security and information access. The proposed architecture could be used while utilizing Retrieval-Augmented Generation (RAG) and fine tuning of Mixture of experts models (MoE). It could be used only with RAG, or only with MoE or with both of them. Using roles and security clearance level of the user, documents in RAG and experts in MoE are filtered. This way information leakage is prevented.

A Simple Architecture for Enterprise Large Language Model Applications based on Role based security and Clearance Levels using Retrieval-Augmented Generation or Mixture of Experts

TL;DR

The paper addresses the risk of information leakage in enterprise LLM applications by introducing a simple role-based security architecture tied to NATO-style clearance levels. The approach filters access at both data retrieval (RAG) and model-expert (MoE) stages, using mappings between users, roles, clearance, and documents. It covers training strategies for local open-source LLMs using MoE and describes how inference can be restricted with RAG, MoE, or a hybrid of both. The proposed architecture aims to enable secure, configurable LLM deployments in enterprise environments with minimal architectural complexity and compatibility with existing security practices.

Abstract

This study proposes a simple architecture for Enterprise application for Large Language Models (LLMs) for role based security and NATO clearance levels. Our proposal aims to address the limitations of current LLMs in handling security and information access. The proposed architecture could be used while utilizing Retrieval-Augmented Generation (RAG) and fine tuning of Mixture of experts models (MoE). It could be used only with RAG, or only with MoE or with both of them. Using roles and security clearance level of the user, documents in RAG and experts in MoE are filtered. This way information leakage is prevented.
Paper Structure (15 sections, 6 figures, 2 tables)

This paper contains 15 sections, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background Information
  3. NATO Clearance Levels
  4. Role-Based Access Control (RBAC)
  5. Introduction to large language models
  6. Training of large language models
  7. Retrieval-Augmented Generation (RAG)
  8. Mixture of experts
  9. Proposed Architecture based on Role based security and Clearance Levels
  10. User, Role, Security Clearance to Documents Mapping
  11. Training LLM using Roles and Security Clearance Levels
  12. Inference using Retrieval-Augmented Generation (RAG)
  13. Inference using Mixture of experts models (MoE)
  14. Inference using both RAG and MoE
  15. Conclusion

Figures (6)

  • Figure 1: Example Role Based Access
  • Figure 2: Retrieval-Augmented Generation (RAG) workflow
  • Figure 3: Entity Relationship Diagram for User to Role and Clearance Level to Document Mapping
  • Figure 4: Sequence Diagram for Role/Clearance level based access to LLM only using RAG
  • Figure 5: Sequence Diagram for Role/Clearance level based access to LLM only using MoE
  • ...and 1 more figures