AI's patriarchal grammar: How large language models distort feminist legal analysis

Abstract

This article argues that AI language models can produce what we term "patriarchal grammar:" a recurring distortion of feminist legal scholarship through the euphemization of sexual violence, the erasure of structural feminist critique, and the naturalization of patriarchal assumptions as neutral facts. These distortions pose a serious epistemic risk to feminist legal analysis, particularly in postcolonial Asian contexts where English mediates legal discourse. Using ChatGPT, Claude, and Gemini on the 2022 Indian marital rape case (RIT Foundation v. Union of India), we observed that in the selected encounters documented here, these models transformed "marital rape" into "intimate disputes," reduced Justice Shakdher's feminist constitutional reasoning to generic rights discourse and presented Justice Hari Shankar's patriarchal logic as objective fact. These encounters are presented with sufficient contextual detail to enable critical evaluation, though full replicability is constrained by the instability of hosted commercial AI systems. Following Gayatri Spivak, we ask for whom we are making language accessible, and to what effect. While AI tools promise linguistic equity, they risk enacting epistemic violence by selectively filtering feminist analysis through dominant patriarchal frameworks. We conclude by proposing protocols for feminist scholars to preserve epistemological authority when engaging with AI tools.