What happened
Published 27 May 2026, Trust Imperative 5.0 is the fifth annual installment in BCG's ongoing collaboration with Salesforce on AI governance in government. The report examines whether AI risk, assurance, and governance frameworks are fit for purpose across ten countries and finds a persistent execution gap: most governments have established AI ethics principles and national frameworks, but struggle with day-to-day application. Common uncertainties persist around how to classify AI risk, what evidence is sufficient for assurance, who holds decision authority, and what constitutes 'safe enough' for different use cases. The report adds new urgency because 'current frameworks predate the rise of agentic AI systems, creating unresolved questions about how to manage human oversight, risk, and accountability.' Eight practical priorities are identified — including making risk triage proportionate, embedding assurance into delivery rather than bolting it on, and redesigning assurance for GenAI and agentic systems. The report estimates governments that make assurance more usable can capture up to '$1.75 trillion annually by 2033' in productivity gains while maintaining public trust.
Why it matters
Government CIOs, CDOs, and AI policy leads who assumed national AI frameworks were sufficient will find this report a direct challenge to that assumption — and a practical playbook for closing the gap between principle and practice before agentic deployments outpace existing controls.
Action needed
Use the report's eight priority areas as a diagnostic checklist in your next AI governance review; pay particular attention to the agentic-system assurance section if your organisation is piloting or procuring agent-based workflows in 2026.