刻板印象固化與社會群體貶抑

Risk Description

Social stereotypes in training data are learned and amplified by models, causing AI systems to automatically associate specific groups with negative characteristics when generating content. For example, binding specific occupations with gender, or associating certain ethnicities with crime. Beyond biased output, large-scale use further entrenches social prejudice, creating negative feedback loops.