paper-review llm-security
Paper Review: AgentDojo and the Problem of Evaluating Agents Under Attack
AgentDojo is the best framework we have for evaluating prompt injection attacks on LLM agents. Its most important finding is also its most unsettling one.
Read More →