According to ISA/IEC 42001, AI bias enters at multiple points: 1. Historical bias (training data reflects past discrimination), 2. Measurement bias (proxies correlating with protected characteristics), 3. Aggregation bias (works well on average, fails for subgroups), 4. Deployment bias (used outside the intended context). Bias can be invisible in overall metrics while causing severe harm to specific groups. Disaggregated testing (across demographic groups)