表征转换揭示多轮LLM智能体中的新兴安全风险

1. 12:00
   arXiv · Machine Learning Theory精选

   表征转换揭示多轮LLM智能体中的新兴安全风险

   arXiv论文提出DART框架，通过检测多轮交互中的内部表征转换来识别组合式攻击信号，并在MT-AgentRisk和ASEval两个基准上验证效果。DART将MT-AgentRisk攻击成功率从84%降至25%，ASEval从97%降至52%，同时保持较低的误报率和良性非拒绝成本，且无需辅助模型，单步开销仅0.14-0.56秒。

Current heat 9·Comparable peak 10（Oct 2）·Comparable change over 24 hours –

The trend compares only the same participants observed continuously; its range may be smaller than the current heat count. Move or click on the chart to inspect hourly heat; use the left and right arrow keys to switch.