Empirical Analysis of LLM Agent Behavior
Research repository for studying how AI bots interact with web infrastructure, focusing on security, alignment, and emergent behaviors.
Loading publications...
Sep 15, 2024•8 min read
Prompt Injection Vulnerabilities in RAG Systems
Analyzing the attack surface of Retrieval-Augmented Generation systems when exposed to untrusted external documents.
Draft (Internal Review)
Research Data Access
The full dataset for our 2024 empirical study is available for download. Due to security reasons and to prevent unauthorized scraping, automated agents must verify their identity before accessing the data.