Sovereign-grade security, by default.
BIBA.AI is built for clients who treat intelligence as a national asset. Six pillars define our security posture.
Multi-tenant by design
Every row in every table is keyed by tenant_id and protected by Postgres Row-Level Security. No cross-tenant data leakage is possible at the database layer.
AES-256 at rest
All sensitive intelligence data is encrypted at rest in Supabase Postgres with AES-256. Backups are encrypted in transit and at rest.
Audit hash chain
Every classification, sanitisation, export, and access event is written to an immutable hash-chained audit log. Tampering with historical records is detectable in O(1).
Classification + portion marking
Documents and chunks carry NATO-style classification banners (UNCLASSIFIED → TOP SECRET) and per-paragraph portion marks. Exports respect the highest mark.
Per-user privacy
Conversations, API keys, and personal settings are stored per-user and namespaced in the browser. Sharing a workstation does not leak chat history between accounts.
No client-side AI keys
All Anthropic Claude calls route through a server-side proxy with body-size caps, per-user rate limits, and structured request IDs. Keys never reach the browser.