Security Model

Every AI action passes multi-layer authorization. Permissions renew per policy epoch, retrieval planning follows least privilege, and input screening plus output moderation guard both ends of the pipeline.

If one layer falls, the next holds. That is a design premise, not a hope.