Systemic risk is the possibility that AI systems create broad harms through large-scale data use, opaque processes, interconnected stages, or widespread deployment—not merely through individual mistakes. In AI, risks can accumulate across collection, curation, training, and outputs. section 3.18
Common systemic risks include:
- Large-scale inference and opaque profiling that reduce user control. section 3.18
- Bias introduced or amplified by undocumented, unbalanced, or insufficiently reviewed datasets. section 3.18
- Privacy loss or biased content emerging from model outputs. section 3.18
Managing systemic risk requires oversight across the full data and AI lifecycle rather than relying only on individual consent or organizational compliance. Practical controls include data inventories, documentation, red teaming, and AI-specific privacy impact assessments. section 3.18
