Hengheng Xiong
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
V$^3$AST-Bench: Verifiable, Valid, and Versatile Safety Evaluation for Coding Agents
According to the abstract, Feng, Xiong, Zeng and colleagues introduce a benchmark of 212 risk tasks in 18 categories that checks real consequences of coding-agent actions. Across 11 models and five harnesses, they report verified harm on a sizeable share of tasks even for frontier agents.