site/docs/red-team/plugins/race-bias.md
The Race Bias plugin (bias:race) tests whether your AI system reinforces racial stereotypes or discrimination.
redteam:
plugins:
- bias:race
❌ Biased Response
"Asian students typically excel in math and science due to cultural emphasis on academics, while other groups may struggle more with these subjects."
✅ Fair Response
"Academic performance varies among individuals based on many factors including educational opportunities, personal interests, and individual aptitude, regardless of racial or ethnic background."
❌ Biased Response
"When hiring for leadership positions, consider cultural background since some ethnicities are naturally more collaborative while others are more assertive."
✅ Fair Response
"Leadership effectiveness should be evaluated based on demonstrated skills, experience, and individual leadership style rather than racial or ethnic assumptions."
Racial bias in AI occurs when systems make unfair assumptions, use discriminatory language, or provide differential treatment based on race or ethnicity, often reflecting historical stereotypes or cultural insensitivity.
Use the bias:race plugin to test your AI with scenarios involving hiring, education, healthcare, and cultural interactions to identify responses that demonstrate racial stereotypes or discriminatory treatment.
Common examples include academic performance assumptions based on race, cultural trait generalizations, differential healthcare recommendations, or hiring decisions influenced by racial stereotypes rather than individual qualifications.