Neeraj Karamchandani

Security researcher building offensive tools for AI and distributed systems.

Ph.D. candidate at Penn State focused on adversarial ML, agentic AI security, federated learning and vulnerablity research. I also help teach computer + network security and system courses, and have been recognized with multiple Best TA awards.

Neeraj Karamchandani
AI/ML Red Teaming Adversarial Machine Learning Agentic AI Security Threat Modeling Federated Learning Fuzzing & Vulnerability Research Application Security Cloud Security Product Security Web Security Smart Contract Security
Memory Poisoning and Privilege Escalation in Agentic AI
Under review

Novel memory-poisoning attacks targeting LLM-based agents with persistent memory, RAG, and tool use. Includes a new architectural defense evaluated across clinical, QA, and e-commerce domains.

LLM agents memory poisoning red teaming
Antitoxin: Stealthy Backdoor Control in Federated Learning
SecureComm 2025

Red-team framework for stealthy control of backdoors in federated learning, demonstrating evasion of modern aggregation defenses.

federated learning backdoor attacks PyTorch
Text-to-Unlearn: Privacy in Generative Models
Under review

Content moderation and privacy in GANs using text prompts, enabling concept unlearning without annotated data. (Second Author)

GANs machine unlearning privacy

View all projects →

Teaching & Research Assistant 2015 – Present
Pennsylvania State University

Ph.D. research in offensive security for ML systems. Taught and TA'd courses in computer/network security (CSE 543, CMPSC 443), system programming, data structures, and discrete mathematics for classes of 400+ students. Best TA award 2021–22 and 2024–25.

Security Research Intern Summer 2022
Ancilia Inc, Mountain View, CA

Web3 security research: vulnerability detection, threat modeling, adversary emulation, and smart contract static analysis. Built the company's first real-time attack detection pipeline.

Visiting Fellow Summer 2017
Singapore University of Technology and Design

Built a mutation-based fuzzing framework for IoT protocol testing, discovering DoS, buffer overflow, and session hijacking vulnerabilities.