Hi! 👋🏼
I’m Sanjana Nambiar, an AI security researcher and engineer working on the reliability, robustness, and safety of large language models and generative AI systems.
My interests include adversarial behavior, trustworthy AI, and the reliability of LLMs. More broadly, I’m curious about understanding language models “under the hood,” including mechanistic interpretability, reasoning behavior, and the reliability of agentic AI systems.
Currently, I work as a Junior Research Scientist at the Center for Interacting Urban Networks (CITIES) Research Institute at NYU Abu Dhabi, under Prof. Azza Abouzied and Prof. Christina Pöpper, where I study misinformation, retrieval vulnerabilities, and adversarial behavior in web-searching LLMs.
Previously, during my undergraduate studies, I worked on LLM jailbreak attacks, alignment benchmarking, and adversarial defense strategies through collaborations across NYU Abu Dhabi, CCSAD, and NYU Tandon. Alongside research, I enjoy building practical AI systems and experimenting with agentic workflows.
News
May 2026: Research on misinformation in web-searching LLMs currently under review.
Feb 2026: Joined the Center for Interacting Urban Networks (CITIES) Research Institute at NYU Abu Dhabi as a Junior Research Scientist, working on Data void exploits, misinformation and LLM vulnerabilities.
Jan 2026: Completed the Technical AI Safety Course by Bluedot Impact, focused on frontier AI risks and AI safety evaluation.
Aug 2025: Joined Trevex as an AI Product Researcher, building GenAI-powered agentic workflows and conversational AI systems.
Jun 2025: Published and presented my first first-author paper, JailFact-Bench: A Comprehensive Analysis of Jailbreak Attacks vs. Hallucinations in LLMs, at SiMLA 2025, co-located with ACNS 2025 in Munich, Germany. [Slides]
May 2025: Graduated from New York University Abu Dhabi with a B.S. in Computer Science and minors in Applied Mathematics and Engineering.
Apr 2025: Co-authored Style Over Substance: Failure Modes of LLM Judges in Alignment Benchmarking, presented at ICLR 2025.
For more, check out Publications, Projects, and CV.