This study challenges the assumption that teachers’ technical competence with Generative AI (GenAI) naturally translates into student adoption. While educators increasingly use GenAI for administrative efficiency and instructional design, student-facing integration remains limited. We conceptualize this tension through the "Pedagogical Firewall": a deliberate exercise of teacher agency aimed at preserving traditional learning hierarchies against perceived AI-related disruptions. Although recent frameworks, such as the DigCompEdu AI Supplement (Bekiaridis & Attwell, 2024), emphasize a progression toward learner empowerment, our findings suggest that such evolution is neither automatic nor linear.

Drawing on a longitudinal mixed-methods study of 20 high school teachers over six months, we analyzed 181 AI-supported activities using the SAMR model (Puentedura, 2006). The analysis identifies a distinct teacher archetype, termed “The Consolidators.” These teachers demonstrated high technical proficiency (DigCompEdu B1/B2), particularly in teacher-facing practices such as generating assessments and rubrics. However, a stark paradox emerged: while technical sophistication in teacher-oriented tasks grew by 22%, the implementation of student-facing GenAI activities increased by a negligible 1%.

Qualitative analysis reveals that this discrepancy reflects a value-based pedagogical stance rather than a lack of knowledge. Teachers often perceive GenAI as a “competitor” to student cognition, positioning it as a tool for efficiency rather than a partner in learning. As one participant noted, AI is viewed as a tool to "teach better," not for students to "learn with." The study argues that advancing AI integration requires shifting professional development toward pedagogical models that support human–AI partnership and redefine teachers’ roles as mentors of co-creation.