Breaking Agents: Compromising Autonomous LLM Agents Through Malfunction Amplification

Open in new window