Observatory · AI governance
Evaluating AI governance: distinguishing management evidence from claims
How to read evidence about the way an organisation governs AI, and why a management system reference does not attest the behaviour of each model.
- Format
- Methodological guide
- Area
- AI governance
- Published
- 31 October 2026
- Author
- International Accreditation Center
- Scope
- General criteria. This guide does not assess any specific organisation, facility or product.
Steps
- Start from uses, not from toolsList where AI influences decisions, who approves its use and what happens when outputs are wrong.
- Separate system and model questionsGovernance evidence shows how risks are managed; model evaluation tests a specific system. Both may be needed.
- Ask for traceable recordsApprovals, risk assessments, monitoring results and incident records allow a conclusion to be checked.
- Check supplier conditionsContracts and documentation from model or data providers define what the organisation can actually control.
- State the limits of a conclusionA conclusion applies to the scope, period and evidence reviewed.