CSCI 460 - AI Security and Adversarial Machine Learning

Credit Hours: 3

This course examines the security of AI systems from an adversarial perspective. It focuses on how machine learning models and AI-driven systems can be exploited through manipulation of data, models, and system pipelines. Students will analyze the attack surface of AI systems, including vulnerabilities in data collection, training, inference, and deployment processes. The course covers core adversarial techniques such as poisoning attacks, evasion attacks (adversarial examples), model extraction, and model inversion. It also examines how AI is used in cyber attacks, including automation of phishing, malware generation, and large-scale exploitation. Special emphasis is placed on threat modeling for AI systems and understanding how attackers exploit weaknesses in real-world AI deployments, including modern systems such as large language models (LLMs).

Prerequisites: CSCI 459 - Introduction to Deep Learning