Book

Who gave the AI authority to act? As autonomous AI agents move from generating answers to making decisions, invoking tools, transferring funds, modifying infrastructure, negotiating, and acting on behalf of institutions, a new governance problem emerges: technical permission is not the same as legitimate authority. Governing Autonomous AI Agents introduces the AAGSF framework—a practical model for connecting institutional authority to machine-executable action. The book explains how organizations can distinguish capability, identity, permission, delegation, authority, and accountability; detect authority-state divergence; control revocation and delegation; and enforce four runtime decisions: ALLOW, DENY, ESCALATE, or HOLD. Drawing on cybersecurity, AI governance, Zero Trust, authorization architecture, law, and institutional design, Emmanuel Nabuora Nakitare develops a detailed reference architecture for authority-aware AI systems. The book also examines applications in finance, procurement, healthcare, government, cloud security, critical infrastructure, and space systems, supported by a 500,000-action synthetic evaluation and a practical adoption playbook. For technology leaders, security professionals, policymakers, developers, legal and compliance teams, and researchers, this book offers a central principle for the agentic era: A machine’s executable authority should never exceed the legitimate authority from which it derives.