Gender Bias in Generative Artificial Intelligence: Genealogies of Inequality, Technological Reproduction, and Feminist Futures

Gender bias in generative AI is an inextricably socio-technical phenomenon that requires a unified perspective across computer science, feminist epistemology, and governance research rather than technical or social critique alone. Analyzing this issue across five distinct levels—from historical data production and model mechanics to benchmark testing and real-world deployment—the study demonstrates that standard fairness metrics vary significantly across tasks, languages, and prompts. Consequently, quantitative test results cannot be evaluated in a vacuum; they represent situated measurements that demand theoretical interpretation rather than simple numerical solutions. Furthermore, because technical mitigations alone fail to fix underlying structural inequalities, the proper focus of analysis must be the "model-in-context"—how the AI operates within specific institutional workflows such as healthcare, education, and employment. To address these systemic limits, the authors propose a comprehensive, multilevel governance model. This framework combines technical evaluations, transparent documentation, community oversight, and public-interest knowledge infrastructure, arguing that democratic and feminist governance is essential to keep alternative, equitable technological futures open.