Research

My research focuses on offensive security in machine learning, with the goal of exposing vulnerabilities in AI systems before they are deployed at scale. I work at the intersection of adversarial ML, agentic AI security, and federated learning.

Publications

Antitoxin: Stealthy Backdoor Control in Federated Learning

SecureComm 2025

A red-team framework for controlled insertion and removal of persistent backdoors in federated learning. Demonstrates evasion of modern aggregation defenses and challenges existing FL threat models. Achieves near-total backdoor eradication in 26 epochs across CIFAR-10, CIFAR-100, and EMNIST.

SMuF: State-Machine Mutational Fuzzing for IoT

ICCIS 2018 (Springer)

A mutation-based fuzzing framework leveraging state machine models and packet mutation to discover DoS, buffer overflow, and session hijacking vulnerabilities in IoT protocol implementations.

Under Review

Memory Poisoning and Privilege Escalation in Agentic AI Systems

Top-tier security venue

Novel memory-poisoning attacks against LLM-based agents with persistent memory, RAG, and tool use. Proposes SENTINEL, an architectural defense, evaluated across clinical EHR, question-answering, and e-commerce domains with multiple LLM backends.

Text-to-Unlearn: A Cross-Modal Approach to Unlearning in GANs

Security conference

A framework enabling selective unlearning in GANs focused on human faces, using text prompts instead of annotated data. Supports expression and multi-attribute unlearning with automatic evaluation using image-text alignment metrics.