A new study from the University of Washington found that artificial intelligence systems used for resume screening show strong racial and gender bias, overwhelmingly favoring white male candidates. Researchers tested three open-source AI models across more than three million job, race, and gender combinations and discovered the systems preferred resumes with white-associated names 85% of the time, while Black men were almost never selected.
To isolate bias, researchers altered only the names on identical resumes, keeping education and experience the same. Despite this, AI models consistently favored white male candidates — even for roles typically held by women, such as human resources positions. The findings highlight how AI systems can reproduce and amplify existing societal inequalities embedded in their training data.
The study raises serious concerns as employers increasingly rely on automated hiring tools. Researchers warn that simply removing names from resumes is not enough, since AI can still infer identity from factors like education history or language patterns. While some states, including California and New York, have begun introducing laws to address discrimination and transparency in AI hiring systems, experts say addressing bias at the training-data level remains a major unresolved challenge.
