Tool-using language agents increasingly modify external systems, yet most runtime safeguards allocate scrutiny by generic uncertainty scores or coarse mutati…
机构:CMU
来源:arXiv 2610.36570 | AI4Papers 论文推荐平台