Uploaded July 2026 | Updated September 2026, 2 weeks ago
Identifies a core flaw in enterprise RAG: retrieval ranks documents by relevance, not authorization, so one tenant's query can surface another tenant's confidential data. The authors propose a layered isolation architecture combining policy-aware ingestion, retrieval-time gating, and server-side orchestration, validated in the open-source OGX framework where ABAC gating eliminates cross-tenant leakage with negligible overhead.
Speakers & affiliations: Francisco Javier Arceo (Red Hat AI), Varsha Prasad Narsing (Red Hat AI)
Session: Paper Session 5 — Security & Governance · Thursday, May 28
Key terms: enterprise retrieval, multitenancy, access control, RAG security, tool use, data isolation
ACM Digital Library: doi.org/10.1145/3786335.3813145
Identifies a core flaw in enterprise RAG: retrieval ranks documents by relevance, not authorization, so one tenant's query can surface another tenant's confidential data. The authors propose a layered isolation architecture combining policy-aware ingestion, retrieval-time gating, and server-side orchestration, validated in the open-source OGX framework where ABAC gating eliminates cross-tenant leakage with negligible overhead.
Speakers & affiliations: Francisco Javier Arceo (Red Hat AI), Varsha Prasad Narsing (Red Hat AI)
Session: Paper Session 5 — Security & Governance · Thursday, May 28
Key terms: enterprise retrieval, multitenancy, access control, RAG security, tool use, data isolation
ACM Digital Library: doi.org/10.1145/3786335.3813145
