Debeshee Das, AI Security and Privacy Researcher at ETH Zurich, expert in adversarial robustness and privacy-preserving machine learning

AI Security & Privacy Researcher

MSc in Computer Science, ETH Zurich

Hello, I'm Debeshee Das

I'm a second-year Master's student in Computer Science at ETH Zurich, majoring in Machine Intelligence and specializing in AI security and privacy research. My work focuses on understanding and mitigating vulnerabilities in foundation models and AI systems, with particular emphasis on agentic security and privacy-preserving machine learning.

At ETH's SPY Lab, I have worked with Prof. Florian Tramèr and Jie Zhang, demonstrating fundamental flaws in membership inference attack evaluations for foundation models, with findings published at top-tier venues including IEEE Security & Privacy, ICLR Data-FM Workshop, and IEEE SaTML. Currently, I'm developing novel defenses against prompt injection attacks on LLM agents during my summer internship at Snyk/Invariant Labs with Dr. Luca Beurer-Kellner and Dr. Marc Fischer, achieving near-perfect attack reduction while maintaining system utility. I have also completed research internships at Microsoft Research (cloud incident analysis) and IBM Research Labs (AI for software engineering).

Previously at IIT Tirupati, I graduated as the top student across all departments (President's Gold Medal recipient) and have been recognized with prestigious awards including the IEEE WISC Premier Scholarship and Pralhad P. Chhabria Award for excellence in STEM research.

Research Interests

AI Security and Safety
Foundation Models
Prompt Injection Defense
Membership Inference
Adversarial Robustness
Privacy-Preserving ML

Research Experience

Selected Publications

Loading selected publications...

Education

MSc in Computer Science

ETH Zurich

2023 - Present

Major: Machine Intelligence

B.Tech in Computer Science and Engineering

IIT Tirupati

2019 - 2023

CGPA: 9.85/10.0 (1st Rank)

Awards and Fellowships

Loading awards...

hourglass_empty

Loading publications...