<p>As AI systems increasingly generate and distribute messages on behalf of organizations, audiences may judge not only the content they receive but also the fairness of the system behind it. This study examines whether ideological cues from an institutional source shape perceived algorithmic fairness in AI-generated communication. Drawing on the Hostile Media Effect, we conducted an online experiment with U.S. participants (<i>N</i> = 280) in which the same AI-generated public service announcement was attributed to either a liberal or conservative publisher. Although the message content was identical and non-political, publisher ideology shaped audience responses. More importantly, moderated mediation analyses showed that ideological alignment increased perceptions that the AI-generated message reflected a fair, neutral, and unbiased system, whereas ideological misalignment reduced perceived algorithmic fairness and led to less favorable attitudes, willingness to support, and e-WOM intentions. These findings extend hostile media research by showing that source-based bias in AI communication can target the perceived fairness of the AI system itself, not merely the message or publisher. The study contributes to AI ethics and governance by demonstrating that public evaluations of AI-generated outputs depend not only on system properties but also on the institutional contexts through which they are encountered and interpreted, raising ethical concerns about how perceptions of fairness may be shaped by sociopolitical cues rather than the actual behavior of the system.</p>