Large language model agents increasingly take consequential actions in digital and physical environments, where safety risks may emerge only after multiple i…
机构:哈工大
来源:arXiv 2611.02815 | AI4Papers 论文推荐平台