Can Computer Science Instructors See Inside the Cognitive Black Box of Student AI Use?

Generative AI tools such as ChatGPT and Copilot are rapidly reshaping how students approach computing tasks, but far less is known about how instructors interpret these changes in students' reasoning and problem-solving. Building on prior work examining faculty concerns and adaptation to AI tools (Lau & Guo, 2023; Sheard et al., 2024), this study presents early findings from a qualitative, multi-institutional interview study exploring how computer science instructors perceive the cognitive and pedagogical impact of generative AI in their classrooms. Through semi-structured interviews with six experienced instructors from four different institutional contexts, we examine how educators interpret students' evolving mental models, cognitive engagement, and tendencies toward both overreliance and underreflection when using AI. Participants described strategies such as requiring students to document prompts, explain outputs, and credit AI contributions – efforts to increase transparency and make thinking more visible. At the same time, instructors noted students' growing emotional distance from their work, uncritical trust in AI-generated answers, and uncertainty about evaluating ''authentic'' problem solving when AI blurs intention and automation. Designed for computing educators at all levels, this lightning talk seeks discussion and community insight on how to support reflective, equitable, and pedagogically sound student use of generative AI. By surfacing instructors' perspectives, our goal is to move the conversation beyond fear or blind adoption toward curiosity, collaboration, and constructive approaches to navigating the ''cognitive black box'' of student AI use.