AI Assurance Questionnaire (v4.3)
The redrafted AI Assurance Question Bank consolidates the Home Office assurance approach across the full AI lifecycle. Each question is mapped to a project governance stage, a lifecycle stage, a risk tier, and an ethical dimension.
At a glance
How to use this navigator
- Browse Questions — filter the full bank by governance stage, lifecycle stage, risk tier, ethical dimension, or free-text search.
- By Lifecycle — see how assurance shifts across Planning → Development → Testing → Deployment → Monitoring.
- By Governance Stage — expand each of the 10 governance stages to see its questions.
- Ethical Coverage — bar chart of question volume across the 15 ethical dimensions.
- Heatmap — cross-tabulates ethical dimensions against lifecycle stages.
Each question is tagged for proportionality. All applies universally. High is mandatory for high-risk/business-critical systems. GenAI applies specifically to generative AI deployments.
Browse all 146 questions
Combine filters to drill into the bank. All filters are AND-combined.
Questions by AI Lifecycle Stage
Click a stage to see questions mandated at that point. Counts show questions tagged to each stage.
Questions by Project Governance Stage
The questionnaire is organised into 10 governance stages. Expand each to see its full question set.
1. Governance & Accountability22
2. Use-Case & Impact Assessment39
3. Data Governance12
4. Model Development20
5. Validation14
7. Security & Privacy8
8. Performance & Monitoring9
9. Human Oversight7
10. Compliance & Legal8
11. Deployment & Change Control7
Ethical Dimension Coverage
Distribution of the 146 questions across the ethical dimensions, with the coverage assessment from the source workbook.
What “coverage” means
- Strong — comprehensive coverage with multiple questions across the lifecycle.
- Adequate — minimum viable coverage; review whether depth matches risk profile.
- Needs strengthening — gaps identified; future iterations should add questions here.
Ethical Dimensions × Lifecycle Stages
Number of questions at each intersection. Darker cells indicate higher concentration; gaps highlight where assurance attention is thinner at a given lifecycle point.
Most assurance work concentrates in Planning & Design, which is expected for an upfront assurance framework. Empty cells highlight stages where a given ethical dimension is not currently exercised — useful when reviewing whether ongoing monitoring is proportionate to design-time scrutiny.