Definition
A data-governance technique where an AI agent's database queries are enforced against the verified identity of the specific human user who asked the question, rather than a shared service account. The data platform carries the end user's permissions all the way down to the data layer. This solves the classic problem where agent tools run under one powerful role that can see everything.
Why it matters
Without it, any user can see whatever the agent's shared account can reach — the single biggest governance blocker for enterprise AI assistants handling sensitive data.