About
I am an M.Sc. student in Electrical Engineering at Tel Aviv University, advised by Dr. Ben Nassi. My research focuses on the security of multimodal large language model (LLM) applications, with an emphasis on adversarial behavior that arises from non-textual inputs.
Before starting my M.Sc., I completed a B.Sc. in Electrical and Computer Engineering at Ben-Gurion University of the Negev, where I specialized in deep learning, digital signal processing, and computer science.
Research
My current work is on securing multimodal LLM applications against non-textual promptware variants. I study how to build model-agnostic, embedding-based guardrails that detect adversarial prompts across modalities, and how to evaluate the robustness of multimodal systems using embedding-space similarity under structured perturbations.
More broadly, I am interested in:
- Security and safety of LLMs and multimodal AI systems
- Promptware and jailbreak detection beyond plain text
- Adversarial machine learning and robustness benchmarks
- Representation learning and embedding-space analysis
Publications
Background
During my B.Sc. I worked on deep-learning-based approaches to quantization and digital signal processing, which led to the ICASSP 2024 publication above.
Alongside my academic work, I have industry experience as a silicon validation engineer (firmware) at Habana Labs (Intel), working on AI accelerators, embedded firmware, and performance characterization of deep learning workloads. This experience strongly influences the way I think about secure and robust machine learning systems in practice.
Contact
The best way to reach me is by email: itamar.assaf@gmail.com.
You can also find some of my code on GitHub.