Evaluating fairness in ChatGPT

A study found ChatGPT may change replies when a user’s name signals gender or race, even for otherwise identical prompts.

The researchers focused on “first-person fairness,” where bias affects the person using the chatbot directly.

They tested English text interactions using common U.S. name-based associations and found signs of identity-linked differences.

The authors note important limits: binary gender assumptions, four race/ethnicity categories, and text-only analysis.

Because chatbots are used in everyday tasks, even small biases can shape the quality of help users receive.