Role-based access, human approval gates, audit trails, and reversibility — the governance layer that enforces organisational policy on every AI agent action.