Governments have moved rapidly from experimenting with artificial intelligence (AI) to deploying it in operational settings. But a gap has emerged between governments’ ambition to scale AI and their ability to deploy systems with confidence.

This report finds that most governments have established AI ethics principles, governance models or national frameworks. The challenge lies in day-to-day execution.

Across the ten countries examined – including Australia – common uncertainties persist around how to classify AI risk, what evidence is sufficient, who holds decision authority and what constitutes 'safe enough' for different use cases. Where the answers are unclear, assurance contributes to delays, duplication and inconsistent outcomes.

This report assesses whether AI risk, assurance and governance frameworks are fit for purpose, how they perform in practice, and whether they strike the right balance between creating public value and mitigating risk. It also highlights what is working well in AI assurance, where friction arises and where targeted refinements could better support AI adoption at scale. Governments that make assurance more usable can capture significant productivity gains – up to $1.75 trillion annually by 2033 – while maintaining public trust.

Key findings

• Ambiguous standards and fragmented processes frequently result in delays, duplication and overgovernance, particularly for lower-risk productivity use cases.
• The emergence of today’s advanced generative and agentic AI systems introduce new questions about delegated authority, human accountability and shared responsibility across agencies, providers and platforms.
• Governments can improve outcomes by making risk triage easier, clarifying accountability, embedding assurance into delivery processes, strengthening internal capability, creating reusable artifacts and redesigning assurance for agentic systems.