This study examines how lecturers and students discursively construct generative artificial intelligence (GenAI) as either a threat or a partner in higher education, and how these competing stances are legitimized in everyday academic communication. Guided by Critical Discourse Analysis, the study analyzes 124 GenAI-relevant statements (74 spoken; 50 written) drawn from assignment briefings, classroom discussions, consultation meetings, LMS announcements, rubric explanations, emails, and class-group chats in a bounded university context. Polarization was coded as Threat, Partner, or Hybrid/Ambivalent, while legitimation strategies were traced using Theo van Leeuwen's framework (authorization, moral evaluation, rationalization, and mythopoesis), complemented by discursive legitimation perspectives. Findings show that Threat-oriented statements (n=52) slightly outnumber Partner-oriented statements (n=35), with Hybrid "allowed, but…" regulation (n=37) functioning as the dominant mechanism for stabilizing norms under uncertainty. Threat talk occurs during high-stakes assessment moments and is commonly legitimized through authorization (rules, rubrics, sanctions) and moral evaluation (fairness, honesty), often reinforced by cautionary narratives. Partner talk becomes more salient in formative learning episodes and is primarily legitimized through rationalization that frames GenAI as scaffolding for brainstorming, outlining, summarizing, and language polishing, while reasserting student responsibility for accuracy and voice. Across stances, the central legitimacy struggle concerns accountability, who remains responsible for claims, originality, and transparency when GenAI is involved alongside student anxiety produced by inconsistent boundaries across courses. The study argues that sustainable governance requires clearer communicative templates, process-oriented assessment evidence (draft trails, reflective logs, brief oral explanations), and careful use of detection tools only as triggers for follow-up review to protect fairness and trust.