What if the AI is asked something outside what we approved?

A general model will attempt anything. Left ungoverned, an assistant built for benefits questions will produce a confident legal opinion, outside every review process the organization established and with nothing marking it as out of remit.

Scope boundaries are governance-tier skills in Centralpoint, loading before the request is processed so they cannot be talked around by later phrasing. Taxonomy scoping reinforces the boundary structurally: a prompt bound to a branch has nothing outside it to retrieve. Escalation routes through Data Triggers as a workflow event with an owner, and the Interaction Log records that the question was asked and how it was handled.