Reproduce “Securing the Agent: Vendor-Neutral, Multitenant Enterprise Retrieval and Tool Use” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Francisco Javier Arceo, Varsha Prasad Narsing

A vendor-neutral architecture, implemented in the open-source OGX framework by Red Hat AI, enables secure, multitenant agentic AI for enterprise retrieval and tool use. This system eliminates cross-tenant data leakage by integrating Attribute-Based Access Control (ABAC) into the data path and enforcing policies through server-side orchestration, introducing negligible performance overhead.

Key insights

  • The 'relevance-authorization gap' is a fundamental security vulnerability in multitenant RAG, where relevance optimization can override authorization, necessitating explicit policy-driven gating.
  • Centralizing agent orchestration server-side is crucial for robust security in enterprise multitenant environments, as it prevents malicious client bypasses and ensures auditable policy enforcement.
  • Policy-aware data ingestion and fine-grained retrieval-time authorization are essential for maintaining strict tenant isolation without significantly impacting performance for shared LLM inference.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.