aka Rahul Deivasigamani
Exploring the security of AI systems, from the prompts they read to the traces they leave.
How do we make intelligent systems more secure?
I’m a master’s student in Cybersecurity & AI at Radboud University in Nijmegen, the Netherlands, and a research intern with its Digital Security Group (DiS). My interests lie at the intersection of security and machine learning: AI-driven threats, the vulnerabilities of large language models, and adversarial robustness.
My research explores indirect prompt injection in mobile AI agents and the robustness of LLM watermarking against adversarial attacks. Before moving into research, I spent four years at Deloitte Consulting USI developing and maintaining enterprise SAP systems.
Off the clock, I enjoy cooking, gaming on my PlayStation 5, and backpacking through new cities.
Our paper, “Not an A11y”, was accepted to AGENT-SEC @ ACM CCS 2026.
Joined the Digital Security Group at Radboud University as a research intern, working on LLM watermarking security.
Started my MSc in Cybersecurity & AI at Radboud University.
Investigating how Android accessibility interfaces expose mobile AI agents to indirect prompt injection.
Read the preprintSecurity research & machine learning.
Designed an attack taxonomy and benchmark across nine Android accessibility attack pathways for Mobilerun and Mobile-Use. Experiments achieved attack success rates of 82% on Gemma4 and 54% on Qwen3.6.
Related publicationBuilt a dual-architecture translation pipeline with byte-level ByT5 and a 9B-parameter Tower+ model fine-tuned using PEFT/QLoRA. Developed memory-efficient inference across dual T4 GPUs using CPU offloading and BF16 mixed precision.
AUG 2025 — PRESENT
Radboud University · Nijmegen, Netherlands
GPA: 8.27/10
Coursework: Security and Privacy of Machine Learning, Deep Learning, Online Tracking & Privacy, Applied Cryptography, and Machine Learning in Practice.
AUG 2017 — JUN 2021
BMS College of Engineering · Bangalore, India
GPA: 8.06/10
MAR 2017
Indian Public School · Kuwait
GPA: 8.42/10
JUL 2026 — PRESENT
Digital Security Group (DiS), Radboud University
Nijmegen, Netherlands
Researching the security of watermarking techniques used in large language models, focusing on their robustness against a novel substitution-based adversarial attack.
AUG 2021 — AUG 2025
Deloitte Consulting USI
Bangalore, India
08 / GET IN TOUCH
I’m interested in research-focused opportunities at the intersection of security and AI.
deiv.rahul@gmail.com