Wei Zou

AgentWatcher is a novel, rule-based monitor designed to detect prompt injection attacks in LLM agents by focusing detection on causally influential context segments, thereby improving scalability and explainability.

The paper introduces eTAMP, a novel attack that poisons LLM web agents' memory using only environmental observations, demonstrating cross-site and cross-session compromise without direct memory access.

CleanBase is a method that detects malicious documents in RAG knowledge databases by identifying clusters (cliques) of documents that exhibit unusually high semantic similarity.

The paper introduces History-Bootstrapped Flow Matching (HB-ARFM) to solve ill-posed spatiotemporal inverse problems, enabling the reconstruction of full physical fields from partial observations by leveraging the observation history.