51.516 Trustworthy AI: Designing Robust, Ethical, and Secure AI System

Course description

This advanced elective explores how AI systems fail under adversarial conditions, and how to design them to be robust, fair, explainable, and secure. Students learn core theory and methods across adversarial machine learning, robustness evaluation, fairness assessment, and assurance approaches such as testing and verification. The course balances research depth with practical system design, including governance considerations and real-world case studies. Hands-on labs use state-of-the-art open-source toolkits such as IBM ART and Microsoft Counterfit, alongside relevant benchmarking and evaluation frameworks. Design scenarios may include biometric spoofing, LLM robustness and prompt-based attacks, and AI supply chain risks. The module culminates in a Trustworthy AI System Design Project where students propose, justify, and defend an end-to-end trust framework for a selected AI application.

Instructor

Victor Keong, Shiva Venkatraman

Number of credits: 12