Bias & Fairness
Identifies discriminatory framing and demographic stereotyping, and records every verdict for regulatory transparency.
What it detects
The Bias pillar evaluates whether a response treats people or groups unfairly — through stereotyping, loaded framing, unequal assumptions, or discriminatory recommendations.
Protected dimensions
- Race, ethnicity, and nationality
- Gender and gender identity
- Age, disability, and religion
- Sexual orientation and socioeconomic status
Flags
| Flag | Meaning |
|---|---|
stereotyping | Generalized assumptions about a group. |
discriminatory_framing | Unequal or prejudicial framing of a topic. |
Why it matters for compliance
For hiring, lending, housing, and insurance use cases, demonstrable fairness controls are increasingly required (EEOC, EU AI Act). Every bias verdict is written to the audit trail, giving you defensible evidence that fairness was actively checked on each output.
Was this page helpful?