Abstract
<p>The global disruption of performance assessment by Generative AI has made reactive detection approaches insufficient. This paper proposes 'AI-Resistant Tasks'—learning activities whose structure prevents LLM completion without meaningful human cognitive engagement. Drawing on the IAT model, this study positions teachers as 'Bilingual Learning Designers' who cultivate 'Reverse PCK' to redesign assessment practices. Analysis of 263 tasks from 138 teachers reveals that experienced teachers created tasks with notably lower vulnerability rates compared to novices (8.7% vs. 23.0%). The findings identified four mechanisms underlying resistance: contextual anchoring, complex role-playing, metacognitive reflection, and process documentation. This study offers an internationally scalable framework that shifts assessment from AI detection to pedagogical empowerment, restoring teacher agency in the GenAI era.</p>