1
Plant, Persist, Trigger: Sleeper Attack on Large Language Model Agents
揭秘针对大型语言模型代理的“潜伏者攻击”：植入、持久化、触发三阶段，AI安全新威胁！
arXiv:2605.28201v1 Announce Type: new Abstract: Large Language Model (LLM) agents remain vulnerable to safety threats from the external environment, w…