LLM08: Vector and Embedding Weaknesses

OWASP LLM Top 10 (2025)

Vector stores and RAG pipelines leak or contaminate retrieved context.

What this risk means

RAG systems are only as secure as the retrieval and embedding pipeline. Misconfigured access control on vector stores, cross-tenant retrieval, or embedding inversion can leak entire knowledge bases. Vector-DB vendors and the customers using them share this surface.

How TrustAtlas dimensions address it

Data-handling covers per-tenant isolation, encryption, and retention; security covers access-control posture (RBAC, audit logging, MFA on admin paths).

Data handlingSecurity

See methodology for how each dimension is scored across the catalog.

Questions to ask vendors

Drop these into RFPs, due-diligence questionnaires, or a procurement scorecard. Each question maps back to evidence visible on the vendor's TrustAtlas profile.

  1. How is per-tenant isolation enforced in your vector store — separate namespaces, separate clusters, or schema-level row filters?
  2. Are embeddings encrypted at rest? Is customer-managed-key (BYOK / CMK) available, and on what tier?
  3. What is your access-control posture on admin paths — is MFA enforced, and is privileged access logged immutably?
  4. Have you evaluated your embedding model against embedding-inversion attacks, and can you share that evaluation?
← LLM07: System Prompt Leakage LLM09: Misinformation →

Related