Matt Fredrikson

Matt Fredrikson is an Associate Professor in the Computer Science Department at Carnegie Mellon University, affiliated with CyLab and the Principles of Programming Group. His research bridges security, privacy, and formal methods in machine learning and software systems.

• PhD in Computer Science, University of Wisconsin–Madison (2015)
• M.S. in Computer Science, University of Wisconsin–Madison (2012)
• Bachelor's in Mathematics and Computer Science, Duquesne University (2007)

His work focuses on privacy in machine learning, particularly adversarial inference and differential privacy limitations. He develops formal methods for privacy-aware programming, using logics with counting to model adversarial uncertainty. Additionally, he explores probabilistic program analysis to enhance machine learning security and reliability.

Recent publications highlight his contributions to LLM security, including attacks on alignment and robustness certification. His 2025 paper LLM Whisperer reveals biases in LLM responses, while 2024 works address certifiable robustness and automated adversarial attacks on coding models.

• Best Paper Award, USENIX Security Symposium 2014

He advises students on topics spanning AI ethics, program verification, and IoT security. Courses taught include Software Foundations of Security and Privacy and Bug Catching: Automated Program Verification and Testing.