7 AI Lifecycle Operation › 7.2 Responsible Development › 7.2.2 Trustworthiness Characteristics
7.2.2.1 Fairness & Bias Management
Detect, measure and mitigate unfair bias in AI systems across the life cycle.
ISO/IEC 42001 Annex B · B.6.1.2
Implementation
- Define fairness objectives and metrics per use case
- Assess training data for representativeness and bias
- Test models for disparate outcomes across groups
Common pitfalls
- Using a single fairness metric for all contexts
- Stopping fairness testing after launch
Success metrics
- Bias testing >95% of high-risk models tested
- Monitoring fairness reviewed at least every 3 months
Maturity
Current 3 · Target 4