The Accountable Enterprise: An Operating Model for Agentic AI in Regulated Institutions

AI agents now take actions in an institution’s name, often without anyone checking each step. Authority can be delegated to software; accountability cannot. Existing frameworks govern models, systems and roles, but none governs the grant of authority itself. This paper proposes the Mandate Model: no agent acts in the institution’s name without a current, written mandate, built on five decisions a board takes once as policy: ownership, authority, evidence, reversal and renewal. It provides an autonomy ladder with a materiality axis, a gap analysis of ten frameworks, a 25-item scorecard with gating items, a 90-day implementation plan, and mappings to QCB, SAMA and CBUAE requirements for regulated institutions in the Gulf.