Nirav Diwan

PhD @ UIUC

formal_profile.JPG

I am a Ph.D. student in Computer Science at the University of Illinois Urbana-Champaign, advised by Prof. Gang Wang in the Siebel School of Computing and Data Science. I also collaborate with Prof. Varun Chandrasekaran and Prof. Huan Zhang. I am interested in security risks of foundation model training and deployment.

Prior to this, I completed my undergraduate studies from IIIT-Delhi, where I had the good fortune of working with Prof. Tanmoy Chakraborty (now at IIT Delhi), and Prof. Zubair Shafiq (at UC Davis), and Prof. Ganesh Bagler.

Internship. I am looking for both industrial and academic internships in Machine Learning, and S&P. Feel free to reach out!

Collaboration. I am always looking to work with undergrads and MS students! Feel free to send me an email with the title [Together] in the subject line!

Industry experience

Prior Research Experience

News

Jul 6, 2026 CoT-Guard has been selected for an oral presentation at AdvML-Frontiers × CoTMA Workshop at COLM 2026! Tweet
May 6, 2026 We release CoT-Guard a 4B model that outperforms larger models (e.g GPT-5.4, GPT5-mini) for CoT Monitoring!
May 1, 2026 Our work on Extractable Memorization from Differentially Private LLM is now accepted at Theory and Practice of Differential Privacy (TPDP) Workshop 2025! See you in Boston!
Apr 1, 2026 Passed my Quals — Officially a PhD candidate now!
Mar 7, 2026 Released a blog post on extracting training data from VaultGemma - the first DP-SGD trained LLM.
Jan 2, 2026 Our simple evaluation study on using LLMs for scraping by everyday users is now accepted at AICS Workshop@AAAI 2025!
Jul 22, 2025 🥇 Our work PurpCode, the first reasoning model for secure code generation developed using Deliberative Alignment, won the Amazon Nova AI Challenge ($250k prize)!
Jul 22, 2025 PurpCode is now accepted at NeurIPS 2025! See you in San Deigo!
Dec 1, 2024 My summer internship work at LG AI Research has been accepted at AAAI Good Data Workshop 2025.
Sep 16, 2024 Our proposal for the Amazon Trusted AI Challenge (Grant - $250,000) got accepted.