Narratives Make AI Misalignment Self‑Fulfill

Public texts and media that describe AIs as hostile or self‑preserving enter model training data and prompt models to imitate those scripts, creating a feedback loop where discourse helps produce the risks people fear. Managing model risk therefore requires not only technical fixes but also attention to how elites, media, and communities talk about AI. — If true, the idea reframes AI safety as partly a social‑narrative problem: what we write and amplify can shape model behaviour and regulatory outcomes.