Sanjana Nambiar
Junior Research Scientist
Affiliation: NYU Abu Dhabi
Education: BSc Computer Science, New York University Abu Dhabi
Research Websites: Center for Interacting Urban Networks (CITIES)
Research Areas: AI Safety; Mechanistic Interpretability; Large Language Models; Agentic AI Security; AI Control; Robustness; Stress-testing AI
Sanjana Nambiar is a Junior Research Scientist at the Center for Interacting Urban Networks (CITIES) Research Institute at NYU Abu Dhabi. She earned her Bachelor of Science in Computer Science from NYU Abu Dhabi, with minors in Applied Mathematics and Engineering. Her academic training spans machine learning, large-scale data systems, computer security, and generative AI, including graduate-level coursework through NYU Courant. Her research is broadly situated in AI safety and robustness, with a focus on understanding failure modes in modern AI systems.
Nambiar's recent and ongoing scholarly work centers on the security, alignment, and evaluation of large language models (LLMs). She is the first author of "JailFact-Bench: A Comprehensive Analysis of Jailbreak Attacks vs. Hallucinations in LLMs," published at the SiMLA 2025 workshop co-located with ACNS, which introduced a benchmark dataset and empirical analysis distinguishing adversarial jailbreak behavior from factual degradation. She has also co-authored work presented at ICLR 2025 examining failure modes of LLM judges in alignment benchmarking, and has conducted research on synthetic fine-tuning as a defense mechanism against privacy and prompt-injection attacks.
Nambiar’s current research interests include mechanistic interpretability, identifying and characterizing sycophantic behavior in AI systems, and stress-testing models under data-void conditions. She is particularly interested in agentic AI security, AI control, and developing principled evaluation frameworks that expose hidden failure modes. Alongside her research, she has participated in and placed at several hackathons and data challenges, and has completed formal training in AI safety through the BlueDot Impact Technical AI Safety Course. Beyond academia, she has contributed to applied AI and data systems through industry roles and interdisciplinary projects spanning urban analytics, cybersecurity, and environmental sensing.