AI in Cybersecurity Education

Faculty Development Summer Institute 2026

Guest Speaker: Malihe Alikhani

[slides]

This morning session focuses on AI alignment as a sociotechnical and institutional relation rather than a purely technical property.

Malihe Alikhani

Malihe Alikhani

Assistant Professor of Computer Science, Khoury College of Computer Sciences, Northeastern University

Dr. Malihe Alikhani is an Assistant Professor at Northeastern University, where she focuses on Natural Language Processing (NLP), AI ethics, discourse and dialogue, and the development of inclusive and equitable AI systems. Her research bridges machine learning, cognitive science, and social sciences to build communicative systems that are safe, fair, and effective. Previously, she was an Assistant Professor in the Department of Computer Science at the University of Pittsburgh and a visiting fellow with the Center on Regulation and Markets at the Brookings Institution. She earned her Ph.D. in Computer Science and a Graduate Certificate in Cognitive Science from Rutgers University.

The Alignment Gap: When AI Agrees Too Much and Institutions Assume Too Little

Abstract:
What does it mean for an AI system to be aligned, and aligned with whom? In this talk, I treat alignment not as a fixed technical property, but as a relation among models, users, tasks, and institutions. I begin with my work on uncertainty and sycophancy in language models, showing how alignment to user intent can collapse into over agreement, misplaced confidence, and reduced epistemic resilience. I then connect these dynamics to AI assisted code generation and productivity, where fast answers are not always useful answers, and where positive friction can help systems clarify assumptions before errors become costly. Finally, I turn to distributional alignment and DPO based methods, arguing that alignment should preserve meaningful variation in human judgment rather than flattening it into a single average preference. I close with brief policy reflections on why institutions should not assume alignment simply because AI systems are fluent, useful, or widely adopted.

Resources