Vitaly Shmatikov

Breaking and Fixing Defenses Against Control Flow Hijacking in Multi-Agent Systems

Harnessing the Universal Geometry of Embeddings

Multi-Agent Systems Execute Arbitrary Malicious Code

Approximating Language Model Training Data from Weights

Rerouting LLM Routers

Soft Prompts Go Hard: Steering Visual Language Models with Hidden Meta-Instructions

Controlled Generation of Natural Adversarial Documents for Stealthy Retrieval Poisoning

Zero-Order Diffusion Guidance for Inverse Problems