Trust Imperative 5.0
B · Useful, limited A qualitative consultancy report based on about 20 interviews across ten countries, jointly funded by BCG and a software vendor (Salesforce), with practical detail but little quantitative evidence.
Summary based on the full text·Original: about 45 minutes to read
In short
The fifth Trust Imperative report (2026) from Boston Consulting Group and Salesforce, developed with the Centre for Public Impact. It examines how governments in ten countries apply AI principles, risk assessments and assurance processes in practice. The finding: principles, accountable roles and frameworks are in place almost everywhere, but they are hard to apply. Risk thresholds are vague, decision rights unclear and reviews duplicated, so low-risk productivity tools get over-governed while questions about agents and shared vendor responsibility stay open.
Key numbers
- 71crossover points where different teams in one agency asked for the same or similar assurance information
- $1.75 trillionBCG estimate of annual GenAI productivity benefits for governments globally by 2033
- ~20practitioner and expert interviews behind the findings
- 10countries whose AI assurance frameworks were analysed
As reported by the source.
English·Web page·opens bcg.com
What it reports
- Countries covered: Australia, Canada, France, Germany, India, Italy, Japan, Singapore, the UK and the US.
- Three governance patterns: federal or devolved (Australia, Canada, US, India); centrally coordinated and tool-oriented (Singapore, with Japan close); layered regulatory (Germany, Italy, France). The UK follows its own model, using existing regulators plus transparency records.
- One Australian agency counted 71 crossover points where teams requested the same or slightly different information on data, finance, ethics and project management.
- Interviewees describe heavy governance of basic, low-risk use cases, valuable cases abandoned because the process looked too hard, and shadow AI where formal routes were slow.
- Frameworks were designed for earlier, more bounded AI; agentic systems raise questions of delegated authority, so governance "starts to look more like" a mix of IT and HR governance.
- For agents, the report assigns accountability across five parties: the agency, the LLM provider, the orchestration platform, the implementation partner and adjacent applications.
- BCG's citizen survey (40+ countries) shows self-reported AI proficiency beyond basic level up by roughly 9% from 2024 to 2026, and weekly AI users up by more than 25%.
- Recommendations: proportionate risk triage, clearer accountability, assurance redesigned for GenAI and agents, assurance embedded in delivery, reusable artifacts, capability building, and measuring whether assurance helps.
How they got it
Desk review of national, subnational and organisational frameworks in the ten countries. This was combined with about 20 interviews with officials, regulators, vendors and experts, then peer and expert review. The authors say the interviews are illustrative and qualitative.
Limits
- Co-funded by Salesforce, which sells agent orchestration platforms; the shared-responsibility model gives platforms a defined role.
- About 20 interviews across ten countries is a thin base, and the anecdotes cannot be generalised.
- The $1.75 trillion figure is a BCG model estimate; the report gives no derivation.
Why it matters for companies
Large companies run the same layered stack of principles, committees and checklists, and they hit the same failure: every function reviews separately and nobody owns the final yes. The 71-crossover example is a cheap diagnostic: count how many times a team must answer the same question before an AI tool goes live. The report's framing of agent governance as partly a workforce question (what authority is delegated, what stays with a person) is directly usable by HR and risk functions.
Summary based on the full text. Read the original for the full argument and data.Read the original on bcg.com
- Tags
- Trust, AI adoption