AI Resonance · 阅读最新日报 · AI 入门推荐

2026-11-09 · America/Los_Angeles · 论文 · #15

Authorization Drift in LLM Agents: Evaluating Boundary Compliance in Multi-Step Tool Use

LLM agents can exceed their authorization boundaries while pursuing user goals without an adversary. We formalize this phenomenon as authorization drift across data, tool, scope, and time, and introduce the Authorization Drift Benchmark (ADB): 648 samples across seven domains, each pairing a sufficient compliant path with an attractive unauthorized shortcut. To ensure executable tasks, tempting shortcuts, and feasible compliant paths, Staged Execution-Reflective Synthesis (SERS) grounds benchmark validation in real agent rollouts rather than LLM self-critique. Across 13 LLMs and three…

Authorization Drift in LLM Agents: Evaluating Boundary Compliance in Multi-Step Tool Use

热度 31.9 / 100;排名与评分保留该期记录。