From AI Assistants to Agentic Public Services: A Practical, Trust-First Architecture for Secure, Accountable and Human-Governed Digital Services

This independent applied research brief examines how public digital services could evolve from conversational AI assistants toward bounded agentic systems capable of executing approved actions while maintaining strong controls over identity, authorization, policy enforcement, security, auditability and human oversight. The report proposes a five-level autonomy model and an Authority Plane that separates AI reasoning from execution permission. Its central principle is that autonomy should increase only as assurance increases, with no state-changing agent action occurring without attributable authority and an independently enforceable permission path. The brief draws on current work from Dubai Future Foundation, Digital Dubai, NIST, the World Economic Forum, OECD, UAE data-protection legislation and relevant technical standards. Proposed architectures and governance patterns are clearly distinguished from existing government implementations or policies. This publication is independent applied research and does not represent, endorse, or claim affiliation with Dubai Future Foundation, Digital Dubai, or the UAE Government.