Sahar Abdelnabi

Sahar Abdelnabi is an AI Security Researcher at Microsoft and will join the ELLIS Institute Tübingen as a Faculty/Principal Investigator. She is co-affiliated with the Max-Planck Institute for Intelligent Systems and Tübingen AI Center, leading the COMPASS Research Group focused on safe, aligned, and steerable AI agents with emphasis on security, human-AI interaction, and cooperative systems.

Her research spans three pillars: (1) Probing AI failures through biases, emergent risks, and misuse scenarios; (2) Developing defenses like white-box control methods and reasoning enhancements; and (3) Leveraging AI for societal good through scientific discovery. Key contributions include coining indirect prompt injection vulnerabilities (2023), pioneering generative AI watermarking (2020), and receiving the ACL2025 Best Paper Award for work on LLM sampling heuristics.

• PhD in Computer Science (2019-2024) from CISPA Helmholtz Center, advised by Prof. Dr. Mario Fritz
• MSc in Computer Science from Saarland University

Research Highlights

Her work bridges AI security and safety with sociopolitical implications, focusing on prompt injection, cooperative multi-agent systems, and contextual integrity. She has been recognized by policymakers and industry leaders, including NIST, OWASP, and Microsoft's AI Bug Bounty Program.

Scientific Awards

• Best Paper Award at ACL2025
• Best Paper Award at AISec'23 Workshop
• Spotlight Paper at NeurIPS Datasets and Benchmarks 2024

Academic Leadership

She actively contributes to the AI and security communities through:

• Program Committee: IEEE S&P (2026), SaTML (2024-2026), USENIX Security (2025)
• Organized IEEE SaTML'25 LLMail-Inject Challenge
• Reviewed for top conferences: ICLR, NeurIPS, CVPR