A harm taxonomy is a structured classification of the potential negative impacts arising from a technology, used to organise risk assessment and mitigation. In AI it categorises harms such as misinformation, discrimination, privacy violation, manipulation, and physical or economic damage. A clear taxonomy enables systematic red-teaming, policy mapping, and accountability for deployed systems.
Content
- Taxonomies typically separate harms by affected party, severity, and causal mechanism, distinguishing representational from allocative harms and individual from societal effects. They give governance teams a shared vocabulary to map mitigations, regulatory obligations, and evaluation tests onto concrete categories.