Enterprise AI is evolving into an operating system where autonomous agents can plan, reason, use memory, invoke tools, run workflows, and collaborate with peers. This evolution creates a new governance problem: existing authorization, security, guardrails, and compliance mechanisms are fragmented and not designed for a unified autonomous AI system.\ \ The paper introduces the Unified Policy Architecture (UPA) as a governance framework for Enterprise AI Operating Systems. UPA offers a single policy model that governs AI agents, tools, workflows, memory, enterprise resources, agent‑to‑agent interactions, and business rules. It expands policy control beyond simple authorization to include runtime obligations, human approvals, compliance checks, audit evidence, and governance evaluation.\ \ Key components of UPA are:\
- a declarative policy language foundation that can describe diverse entities uniformly;\
- policy evaluation semantics that ensure accurate runtime enforcement;\
- extensible plugins allowing custom validators or external decision services;\
- industry policy packs providing pre‑built policies for common compliance regimes;\
- an enterprise governance assessment framework to measure coverage and effectiveness.\ \ The paper also identifies extensions such as multi‑agent coordination policies, provenance‑aware decision making, and stateful runtime governance mechanisms. UPA thus lays the groundwork for building secure, accountable, and governable autonomous AI operating systems.\ \ Review