How can we make an AI system useful while limiting what an attacker, an unreliable model, or an overpowered agent can make it do? In this course, students attack and secure small AI applications. Moving the boundary of trust outward, ring by ring, students will fool a model, poison its training data, extract what it has memorized, hijack a language application through a document, and lock down an AI agent's tools and permissions — alongside how AI security plays out in the real world, from deepfakes to confidential computing.
Grade distribution data hasn't been published for this course yet. Check back later as more terms become available.
University of California, Santa Cruz
Reviews, study tips, and grade data for CSE 108B at UCSC. How can we make an AI system useful while limiting what an attacker, an unreliable model, or an…
The course page is taking longer to load. This summary is still available.
We could not load the full course details. This summary is still available.
Still loading. This can take a moment on a slow connection.
Loading My Class Grades