AI Risk Assessment Methodology Guide
Complete methodology for conducting AI risk assessments including criteria definition, risk identification, analysis, evaluation, and documentation.
Chapter Overview
This chapter provides a complete methodology for conducting AI risk assessments as required by Clause 6.1.2 and Clause 8.2. A robust risk assessment process is fundamental to effective AI governance.
The organization shall define and apply an AI risk assessment process that:
• Establishes and maintains AI risk criteria
• Ensures repeated assessments produce consistent, valid, comparable results
• Identifies AI risks (confidentiality, integrity, availability, and other AI risks)
• Identifies risks throughout the AI system lifecycle
• Analyzes risks (likelihood and consequence)
• Evaluates risks against criteria and prioritizes for treatment
Risk Assessment Process Overview
| Phase | Activities | Outputs |
|---|---|---|
| 1. Establish Context | Define scope, criteria, methodology | Risk criteria document |
| 2. Risk Identification | Identify risks across all categories | Risk list |
| 3. Risk Analysis | Assess likelihood and consequence | Analyzed risks |
| 4. Risk Evaluation | Compare against criteria, prioritize | Prioritized risk register |
| 5. Documentation | Document assessment and results | Risk assessment report |
Phase 1: Establish Context
1.1 Define Scope
Clearly define what is being assessed:
- Which AI system(s)
- Which lifecycle stages
- Which business processes
- Which locations/environments
- Assessment boundaries
1.2 Define Risk Criteria
Likelihood Scale
| Level | Rating | Description | Frequency Guide |
|---|---|---|---|
| Rare | 1 | Very unlikely to occur | Less than once per 5 years |
| Unlikely | 2 | Could occur but not expected | Once per 2-5 years |
| Possible | 3 | Might occur | Once per 1-2 years |
| Likely | 4 | Will probably occur | Once per year |
| Almost Certain | 5 | Expected to occur | Multiple times per year |
Consequence Scale
| Level | Rating | Financial | Operational | Reputational | Individual Impact |
|---|---|---|---|---|---|
| Negligible | 1 | <£10K | Minor disruption | No external awareness | No noticeable impact |
| Minor | 2 | £10K-100K | Some disruption | Local awareness | Minor inconvenience |
| Moderate | 3 | £100K-1M | Significant disruption | Regional/industry awareness | Significant negative impact |
| Major | 4 | £1M-10M | Major disruption | National awareness | Serious harm |
| Catastrophic | 5 | >£10M | Business threatening | International awareness | Severe/irreversible harm |
Risk Level Matrix
| Likelihood / Consequence | 1-Negligible | 2-Minor | 3-Moderate | 4-Major | 5-Catastrophic |
|---|---|---|---|---|---|
| 5 - Almost Certain | Medium (5) | Medium (10) | High (15) | Critical (20) | Critical (25) |
| 4 - Likely | Low (4) | Medium (8) | High (12) | High (16) | Critical (20) |
| 3 - Possible | Low (3) | Medium (6) | Medium (9) | High (12) | High (15) |
| 2 - Unlikely | Low (2) | Low (4) | Medium (6) | Medium (8) | Medium (10) |
| 1 - Rare | Low (1) | Low (2) | Low (3) | Low (4) | Medium (5) |
Risk Level Definitions
| Level | Score Range | Response Required |
|---|---|---|
| Critical | 20-25 | Immediate action required; escalate to senior management |
| High | 12-16 | Urgent treatment required; management attention needed |
| Medium | 5-10 | Treatment required; plan and implement controls |
| Low | 1-4 | Accept or treat as resources allow; monitor |
1.3 Define Risk Appetite
"The organization has low appetite for AI risks that could:
• Cause significant harm to individuals
• Result in regulatory non-compliance
• Damage organizational reputation
The organization accepts moderate risk for AI initiatives that:
• Have potential significant business benefit
• Can be monitored and controlled
• Are reversible if issues occur"
Phase 2: Risk Identification
2.1 Risk Identification Methods
| Method | Description | Best For |
|---|---|---|
| Checklist Review | Use Annex C risk sources as checklist | Comprehensive coverage |
| Brainstorming | Team sessions to identify risks | Creative identification |
| Interviews | Discuss risks with stakeholders | Expert knowledge capture |
| Scenario Analysis | "What if" scenarios | Complex risk chains |
| Historical Analysis | Review past incidents | Known risk patterns |
| FMEA | Failure Mode and Effects Analysis | Technical systems |
2.2 AI Risk Categories
Ensure coverage across all categories (reference Annex C):
- Data Risks: Quality, bias, privacy, provenance
- Model Risks: Accuracy, robustness, explainability, drift
- Technical Risks: Security, availability, integration
- Human Risks: Misuse, over-reliance, skill gaps
- Organizational Risks: Governance, resources, communication
- External Risks: Regulatory, threat actors, technology change
- Ethical Risks: Fairness, transparency, human rights
- Impact Risks: Individual harm, societal harm
2.3 Lifecycle Coverage
Identify risks at each lifecycle stage:
| Stage | Example Risks |
|---|---|
| Design | Unclear requirements, ethical issues not identified |
| Data Collection | Biased data, privacy violations, insufficient data |
| Development | Model errors, security vulnerabilities, poor documentation |
| Testing | Inadequate testing, missed edge cases |
| Deployment | Integration failures, user readiness gaps |
| Operation | Misuse, performance issues, incidents |
| Monitoring | Drift undetected, alert fatigue |
| Retirement | Data retention issues, knowledge loss |
Phase 3: Risk Analysis
3.1 Assess Likelihood
For each identified risk, assess likelihood considering:
- Historical occurrence
- Current control effectiveness
- Threat landscape
- Vulnerability exposure
- Environmental factors
3.2 Assess Consequence
For each identified risk, assess consequence considering:
- Financial impact
- Operational impact
- Reputational impact
- Regulatory/legal impact
- Impact on individuals
- Societal impact
3.3 Calculate Risk Score
Risk Score = Likelihood × Consequence
Document both inherent risk (without controls) and residual risk (with existing controls).
Phase 4: Risk Evaluation
4.1 Compare Against Criteria
- Compare each risk score against risk level matrix
- Identify risks exceeding risk appetite
- Flag risks requiring immediate attention
4.2 Prioritize Risks
Prioritize based on:
- Risk level (Critical → High → Medium → Low)
- Treatment urgency
- Regulatory requirements
- Stakeholder concerns
- Resource availability
4.3 Treatment Decisions
| Risk Level | Typical Decision |
|---|---|
| Critical | Treat immediately or avoid activity |
| High | Treat with priority |
| Medium | Treat according to plan |
| Low | Accept or treat if efficient |
Phase 5: Documentation
AI Risk Register Template
Identification:
• Risk ID (unique identifier)
• AI System (which system)
• Risk Category (data, model, technical, etc.)
• Lifecycle Stage (design, operation, etc.)
• Risk Description (what could happen)
• Risk Source/Cause (why it might happen)
• Affected Parties (who is impacted)
Analysis:
• Likelihood Rating (1-5)
• Consequence Rating (1-5)
• Inherent Risk Score (L × C)
• Existing Controls
• Residual Likelihood
• Residual Consequence
• Residual Risk Score
• Risk Level (Critical/High/Medium/Low)
Treatment:
• Treatment Decision (Accept/Treat/Transfer/Avoid)
• Treatment Actions
• Target Risk Level
• Risk Owner
• Due Date
Monitoring:
• Review Date
• Status
• Notes
Risk Assessment Report
1. Executive Summary
• Scope and objectives
• Key findings
• Critical/high risks summary
• Recommendations
2. Methodology
• Assessment approach
• Risk criteria used
• Team and stakeholders
3. Context
• AI systems assessed
• Scope and boundaries
• Assumptions and limitations
4. Risk Assessment Results
• Summary by category
• Risk register (detailed)
• Risk heat map
5. Conclusions and Recommendations
• Overall risk posture
• Priority treatment areas
• Next steps
6. Appendices
• Detailed risk register
• Risk criteria definitions
• Supporting evidence
Assessment Triggers
Planned:
• Annual comprehensive review
• Before new AI system deployment
Event-Triggered:
• Significant AI system changes
• New AI system development
• AI incidents or near-misses
• Regulatory changes
• Organizational changes
• New risk information
• After corrective actions
1. Define clear risk criteria before assessment
2. Cover all risk categories and lifecycle stages
3. Use consistent methodology for comparable results
4. Assess both likelihood and consequence
5. Document everything - it's mandatory
6. Conduct assessments at planned intervals and when triggered
7. Link risk assessment to risk treatment and SoA