Gustavo Sandoval is an Industry Professor in Computer Science and Engineering at NYU Tandon, where he teaches graduate courses in Deep Learning, Machine Learning, Operating Systems, and Distributed Systems.
His research focuses on AI safety and mechanistic interpretability: understanding how large language models work internally, and using that understanding to make them more reliable and secure. Current work includes why safety behavior degrades during post-training, secure code generation through activation steering, chain-of-thought monitoring, machine unlearning in code models, and interpretability of text diffusion models.
He earned his PhD in Electrical and Computer Engineering from NYU Tandon, advised by Siddharth Garg and Brendan Dolan-Gavitt. Before academia, he spent more than 15 years at Microsoft building products used by hundreds of millions of people, including Office, Hotmail, MSN Mobile, Windows Phone, and GroupMe.
Education
New York University, Tandon School of Engineering
Doctor of Philosophy, Electrical and Computer Engineering
California State University Sacramento 1996
Master of Science, Computer Science
Experience
NYU Tandon School of Engineering
Industry Professor, Computer Science and Engineering
From: 2016 to present
Microsoft Corporation
Principal Engineer and Engineering Manager
From: 2000 - 2015
Patents
Method and system for automatically fitting a graphical display o, (U.S Software Patent: US6456305)
Method and system for generating and displaying a slide show with, (U.S Software Patent: US6396500)
Selected Publications
Suppression is not deletion, ICML 2026 Workshop on Memorization in Foundation Models
Lost at C: A User Study on the Security Implications of Large Language Model Code Assistants, USENIX Security 2023.
Even Heads Fix Odd Errors: Mechanistic Discovery and Surgical Repair in Transformer Attention. 2025.
Adversarial Fine-Tuning for Prompt Injection Defense. 2022.
