GURURisks
- sensitive-document leakage
- answers without sources
- mixed public and internal collections
- unmanaged API keys
- unclear generation cost
Janus.Guru
Data security for document assistants
A technology label does not define a data regime. Each deployment fixes access, storage, model routing, logs, personal data, deletion, and the boundary between public and private surfaces.
Verifiable mechanisms
- separate collection selection
- answers with retrieved sources
- keys limited by collections and quotas
- protected operational surface
- administrative audit logging
Pilot checklist
- documents allowed for indexing
- prohibited answer classes
- collection owner
- source refresh
- data location
- deletion of requests and logs
Frequently asked questions
- How are public and internal data separated?
They should live in separate collections with explicit access rules. Isolation is verified on the deployment, not assumed. - What should happen with sensitive documents?
Do not put them into the public demo. Approve infrastructure, model routing, roles, retention, and deletion first.