Agentic Artificial Intelligence (AI) systems are evolving from passive decision-support tools into autonomous digital workers capable of planning, reasoning, invoking tools, accessing enterprise resources, and executing business actions with varying degrees of autonomy. While enterprises maintain mature governance models for human identities, service accounts, and application identities, existing Identity and Access Management (IAM) frameworks do not adequately address the governance challenges introduced by autonomous agents. Unlike conventional applications, agents dynamically determine execution paths, invoke reusable skills, delegate tasks to peer agents, and perform privileged actions across enterprise systems. These characteristics define a new class of enterprise identity that requires dedicated governance, authorization, and accountability controls. This paper proposes the Enterprise Agent Identity Governance Framework (EAIGF), which treats autonomous agents as first-class enterprise identities subject to lifecycle management, quantitative trust classification, privileged access governance, and end-to-end accountability. The framework's principal constructs form an identity layer comprising the Agent Identity Registry (AIR), the Agent Provisioning Certificate, Agent Trust Tiering driven by an explicit Agent Identity Risk Score (AIRS) formula, Agent Privileged Access Management (Agent-PAM), and a Delegated Trust Chain with Agent Identity Lineage for multi-agent accountability. These constructs are enforced through a policy-mediated runtime boundary that performs capability projection and risk-adaptive authorization. A reference implementation realizing the registry, the AIRS formula, capability projection, and risk-adaptive authorization correctly enforced all ten governance scenarios in the evaluation with zero false-permits on disallowed capabilities and a mean authorization overhead of approximately 0.015 ms per request. A larger randomized evaluation over 600 authorization requests across 80 agents and 25 reusable skills reproduced this result and, against a naive whole-skill authorization baseline, reduced the exposed capability surface by 66.7 percent while blocking every attempted privilege escalation through delegation. A synthetic financial-services walkthrough demonstrates how Identity Governance and Administration (IGA), Privileged Access Management (PAM), and Zero Trust principles extend to agentic environments. EAIGF provides a practical blueprint for governing autonomous agents at scale while preserving least privilege, accountability, and operational resilience.