Bias & Fairness

https://taxonomy.eticas.ai/risk/bias-fairness

Maturity: established

The risk that an AI system produces outcomes that systematically advantage or disadvantage individuals or groups based on protected or sensitive attributes, leading to unequal treatment, reduced accuracy, or unjust impacts. This includes biases introduced through data, model design, or deployment context, and covers both measurable disparities and perceived unfairness in decision-making.

Also known as: Fairness · Bias & Discrimination · Algorithmic fairness

System type: ADM and LLM systems
Lifecycle stages: Pre Processing, In Processing, Post Processing

Risk groups

• Outcome disparities — Risks related to unequal treatment, allocation, or performance outcomes across demographic and protected groups. These are observable differences in what the system does to or for different groups.
• Representational harm — Risks related to how the system represents groups in its outputs — through stereotypes, toxic content, skewed representation, or homogenization. These are harms in the form of meaning and content rather than allocation of resources or opportunities.
• Dynamic & Systemic bias — Risks related to how bias is amplified or perpetuated over time through interaction between AI systems and the social systems they operate in.

Mappings to external frameworks

Standards & frameworks

Taxonomies & vocabularies