AI is more biased than humans when hiring
new research shows llms stereotype applicants more than people do, raising risks as models gain memory and agentic capabilities.
research indicates that large language models frequently form stronger biases against job applicants than human recruiters, contradicting the assumption that ai removes subjectivity. instead of solely inheriting biases from training datasets, the models are developing new stereotypes through their processing of information. this issue becomes increasingly critical as ai companies shift toward agentic architectures designed to remember specific user interactions and details. without specific safeguards, these long-term memory capabilities could allow systems to form and reinforce prejudiced associations, potentially leading to unfair automated hiring practices.