What are common Fairness and Identity failure patterns in AI agents?

Documented Fairness and Identity failure patterns include: Chatbot Bias by Dialect or Name: The Quality Gap.

Fairness and Identity | AI Agent Failure Patterns

> **Goal:** Quality of service does not depend on dialect, name, or other identity cues.

Goal: Quality of service does not depend on dialect, name, or other identity cues.

Language models carry societal associations. The only reliable check is paired testing on your own system, before and after every change.

Shared metrics the patterns below report against. Targets are starting points to tune on your own data, unless a source is given.

Metric	Target	Defined in
Paired-test parity ratio	Within about 0.9 to 1.1 (a heuristic)	dialect and name quality gap (GitHub)
Parity regressions after changes	0	dialect and name quality gap (GitHub)

Pattern	One line
Dialect and name quality gap (GitHub)	Unequal service by dialect, name, or identity cues