2 papers across 2 sessions
A benchmark with realistic security scenarios for web agents based on LLMs