This paper introduces asymmetric visibility as a framework for measuring how AI-generated conflict discourse can make subordinated actors visible while weakening their grammatical agency. Focusing on Palestine, Zionism, anti-Zionism, Iran, United States foreign-policy discourse, and platform moderation, it argues that visibility is not equality when dominant actors retain the grammar of action and subordinated actors appear mainly as risks, victims, crises, or moderation objects. The paper proposes Agency-Preservation Rate and the Asymmetric Visibility Index as tools for auditing agency distribution in AI-mediated political language.