Tenant boundary
Which organization or data domain owns the request?
Back to AI Security
RAG Security Assessment
Check who can retrieve each document before expanding your RAG system.
Review tenant isolation, document permissions and generated responses in retrieval-augmented systems.
Why commission the review
A retrieved passage, citation or generated answer can expose a document to someone who cannot open the source. Review whether your RAG system preserves document permissions through retrieval and response generation.
Best fit: Security, data platform, AI product, and enterprise SaaS teams shipping RAG features over sensitive or multi-tenant data.
Scope
Retrieval architecture, tenant and role authorization, vector indexes and metadata filters, query rewriting, document trust, prompt assembly, injection pathways, source attribution, and response controls.
What you receive
Findings on document access and data exposure.
A map of document access, tests for cross-user and cross-tenant exposure, and findings with fixes for retrieval and response controls.
Timeline
Typically 10 business days from kickoff and access confirmation.
Retrieval authorization path
Carry document permissions through to the answer.
We check permissions at retrieval, prompt assembly and response generation. A document that the requester cannot open should not become an answer or citation they can read.
Authorization chain
1Tenant
2Role
3Source
4Context
5Response
Role boundary
What is this user or service identity allowed to access?
Source boundary
Which document, index, metadata filter, or connector can contribute material?
Context boundary
What retrieved content and embedded instructions enter the model prompt?
Response boundary
Could the answer or citation expose material the requester cannot access?
Assessment detail
The proposal identifies the systems and tests included. The report records findings, unavailable access and any checks that could not be completed.
RAG Isolation Findings
Findings on tenant, role, document, source, and metadata boundaries before retrieved content reaches model context.
Retrieval Boundary Review
Assess how user identity, tenant membership, metadata filters, and query rewriting affect what material can be retrieved.
Source Authorization Matrix
Matrix showing which sources require authorization, attribution, trust labeling, filtering, or exclusion from generation.
Prompt-Injection Resilience Notes
Review how instructions embedded in documents, emails, webpages, and tool output can influence retrieval and response behavior.
Remediation plan for customer review
Prepare a concise technical summary of isolation controls, known limits, and remediation priorities for customer security teams.
What the review covers
Check retrieval and response controls together.
- Designed for RAG systems that combine sensitive content, search, and generated responses
- Covers tenant boundaries, authorization checks, retrieval filters, and response behavior together
- Each finding identifies the affected retrieval or access control and a recommended change
When to use it
Before adding sensitive data or another tenant.
Use this before enterprise security review, customer launch, multi-tenant rollout, or expansion into higher-sensitivity document sets.
Next step