SECURITY / ARCHITECTURE

Security

Explore how an AI trust boundary, approved graph queries and independent business identities keep protected values outside the AI toolset.

WHAT THE ARCHITECTURE REQUIRES

Separation is enforced by infrastructure.

This is a deterministic browser simulation with synthetic data. It illustrates an architecture; the visualization and missing edges do not enforce security. No authentication service, vault or production access-control system is implemented here.

Independent storage and identities

LLM, Agent and Tools form the AI trust boundary. Their only business interface is the restricted Graph Query API. Neither that interface nor the AI toolset has vault credentials, a protected-read tool or a direct or indirect network route to protected values. Authorized business services use independent identities and cannot be invoked by the AI toolset.

Approved metadata only

The query API exposes an explicit set of permitted relationships and attributes. Field existence and relationships also require classification and disclosure approval. References are opaque and do not encode secret values.

Every retrieval surface counts

Production controls must also prevent disclosure through alternate properties, search indexes, logs, embeddings and tool outputs. Demonstration controls change visibility only; they cannot change a production access boundary.