AI Security Atlas › All entries › RAG & Knowledge Retrieval Security

GuardrailFixes medium risks

Protect Embeddings as Sensitive Data

Vector stores inherit source-document classification, access controls, encryption, retention, and audit logging.

Impact

Removes the gap between how documents and their embeddings are protected.

Why this severity

Guardrails take the highest severity of the risks they mitigate: Embeddings Treated as Non-Sensitive (Medium), Embedding Inversion & Membership Inference (Medium).

Method: OWASP Risk Rating Methodology

Related

Mitigates

• Embeddings Treated as Non-Sensitive Vulnerability · Medium severity
• Embedding Inversion & Membership Inference Threat · Medium severity

Builds on

• Document-Level Access Enforcement Guardrail · Fixes high risks

Remediation

1. Apply the same ACLs to vectors as to documents.
2. Encrypt at rest; disable bulk export by default.
3. Log and alert on unusual vector reads.

How to verify

• Access reviews include vector stores.
• Bulk-export attempts by non-admin roles are denied and logged.

Framework mapping

OWASP Top 10 for LLM Applications 2025

• LLM08:2025 Vector and Embedding Weaknesses
• LLM02:2025 Sensitive Information Disclosure

MITRE ATLAS

• AML.M0005 Control Access to AI Models and Data at Rest
• AML.M0019 Control Access to AI Models and Data in Production

NIST AI RMF 1.0

• MEASURE 2.10 Privacy risk of the AI system is examined and documented