Adversarial robustness
Understanding how AI models respond to adversarial inputs and how they can be protected.
Explore this topic ↗Research direction
Securing AI
Protecting models, agents, and AI systems.
Connected questions. New possibilities.
The question behind the work
What does it take to trust AI with real responsibility?
Models and AI agents encounter untrusted inputs and changing environments. We study adversarial attacks, deception, and manipulation, and investigate ways to make AI systems more robust.
Questions we explore.
Agent security
Investigating deception, manipulation, and trust boundaries in AI agents.
Explore this topic ↗A closer look
Ideas taking shape.
In focus
Publication · 2026
Breaking and Defending LLM-Powered Social Media Bot Detection Systems
A study of attacks and defences for language-model systems that identify social media bots.
Explore the story ↗Publication
Publication 2026
Inference Attacks and Defenses on LLM-Powered Phishing Email Detection Systems
Frederik Sherman and Yoni Birman study inference attacks and defenses for LLM-powered phishing email detection. The paper is available through the HumSec 2026 workshop.
Publication
Publication 2026
AdvPIIBench - The Adversarial Collapse of PII Detection in LLM Interactions
Roei Arpaly and Yoni Birman investigate how adversarial prompts challenge the detection of personally identifiable information in LLM interactions.