Vulnerable Agent Lab: attacking a tool-using LLM agent

Building a deliberately vulnerable LangChain agent with file and email tools, then attacking it with indirect prompt injection.

July 5, 2026 · 1 min · 149 words · Lynne Lin