Beyond Hallucination: Evaluating Cultural and Institutional Misinterpretation in Public-Facing LLMs

Public-facing large language models are increasingly used as informal interfaces to institutional knowledge. Citizens, migrants, patients, students, and journalists now ask general-purpose AI systems questions that were previously answered through official websites, public servants, NGOs, or professional intermediaries. Their failures are usually described as hallucination, misinformation, or bias. This paper argues that these categories are necessary but insufficient in multilingual public contexts, where LLMs can also fail interpretively: by misreading institutional authority, flattening jurisdictional nuance, translating legal-administrative categories asymmetrically, or producing unequal guidance across languages. Framing LLMs as interpretive technologies, this position paper proposes interpretive fidelity as an additional evaluation dimension: the capacity of an AI system to preserve institutional meaning, source hierarchy, contextual conditions, epistemic uncertainty, and cross-linguistic consistency when answering questions about public knowledge. We outline a preliminary four-dimensional framework covering factual accuracy, source fidelity, contextual and institutional meaning, and cross-linguistic consistency. We ground the argument in a concrete Brussels case: whether the integration course is mandatory for a non-EU newcomer. This case shows that an LLM can avoid hallucinating a nonexistent law while still giving users different bureaucratic realities in English, French, and Dutch. The failure is not merely factual; it concerns how the model orients the user within a complex multilingual institutional system. Finally, we discuss implications for cultural AI evaluation, including the shift from traditional SEO to Generative Engine Optimization for public institutions. We argue that a positive vision of cultural AI should not only reduce harmful outputs. It should preserve public meaning, support linguistic fairness, and protect human agency in institutionally complex societies.