Public servants are increasingly adopting generative artificial intelligence (GenAI), often without formal oversight. This paper addresses the widening gap between rapid, decentralised GenAI uptake and lagging public governance mechanisms. It focuses on structured experimentation as a critical pathway for governments to build capabilities, manage risks and ensure responsible technology adoption before full-scale deployment. 

Drawing on a systematic review of official guidelines across 14 countries, academic literature and case studies, the paper examines how jurisdictions translate high-level ethical principles into operational practice. The 14 countries reviewed are Australia, Canada, Finland, Ireland, Italy, Japan, New Zealand, Norway, Sweden, Switzerland, the Netherlands the United Kingdom, the United States and Singapore.

The findings reveal that, while guidelines are widespread, they remain highly fragmented and uneven, leading to risk aversion or inconsistent practices. Furthermore, monitoring and evaluation are identified as critical weaknesses, with few governments systematically measuring performance, impact or compliance. To bridge the gap between national strategies and daily public administration, the paper presents an evaluation framework across five core areas: performance, public value, feasibility, usability and risk management.