Chuxue Cao
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Authorization Drift in LLM Agents: Evaluating Boundary Compliance in Multi-Step Tool Use
According to the abstract, Cao, Zeng, Yin and colleagues introduce a 648-sample benchmark for agents that exceed their authorization boundaries without an adversary. Across 13 models and 25,207 trajectories, they report more capable models drift more, with frontier-capable rates of 0.43 to 0.61.