Good practices forcity AI agents A framework for responsibleand people-centric deploymentof AI agents in cities
Contents
Abbreviations and acronyms
Executive Summary
1 Purpose, Audience and Scope
     1.1 Purpose
     1.2 Audience
     1.3 Scope
     1.4 Recommended policy status
2 AI Agents in Government and Cities
     2.1 Working definition
     2.2 How AI agents differ from traditional automation
     2.3 Illustrative city use cases
     2.4 Agent lifecycle
3 Guiding Principles
4 AI Agent Classification and Risk Tiering
     4.1 Autonomy levels
     4.2 Risk tiers
     4.3 Factors that increase risk
     4.4 Uses requiring potential prohibition or exceptional approval
5 Recommended Good Practices for AI Agents
     5.1 Public purpose and policy alignment
     5.2 Scope, mandate and authorized use
     5.3 Governance, ownership and accountability
     5.4 Human oversight and decision authority
     5.5 Legal, ethical and rights-based safeguards
     5.6 Data governance, privacy and confidentiality
     5.7 Security, access control and cyber-resilience
     5.8 System integration and tool-use controls
     5.9 Interoperability and cross-government coordination
     5.10 Testing, validation and assurance
     5.11 Procurement and vendor management
     5.12 Transparency, public communication and trust
     5.13 Public AI literacy and digital inclusion
     5.14 Operational monitoring, auditability and incident management
     5.15 Workforce readiness and change management
     5.16 Performance and public value measurement
     5.17 Lifecycle Management and continuous improvement
     5.18 Intellectual property, training-data provenance and content licensing
6 Pathway to Responsible Autonomous City Capabilities
Annex A: Use Case Intake and Approval Template
Annex B: Risk Classification Matrix
Annex C: Pre-Deployment Readiness Checklist