When AI gets it wrong: False inference and political harm

AI systems are increasingly active agents in political discourse, shaping reputations, narratives, and public perceptions. This commentary examines three real-world cases from Serbia where AI chatbots—Grok and ChatGPT—asserted false claims, spreading false narratives about political collectives or regime-critical individuals. These incidents illustrate how, under the guise of technical neutrality, AI can reinforce dominant narratives, amplify disinformation, and undermine dissent. Drawing on a recently proposed framework for AI regulation (Tomić & Štimac, 2025), we show how failures across three dimensions—decision models, data sourcing, and interface semantics—create pathways for political manipulation and reputational harm. We conclude by reflecting on implications for political deliberation and calling for targeted regulatory and empirical responses.