#artificial intelligence
Apr 2026
AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents
The benchmark evaluates whether agents can recognize and interrupt harm arising from accumulated context, repeated tool use, intermediate actions, and dependencies across steps, and indicates that current systems remain highly vulnerable.
Yun-Hao Feng, Yifan Ding, Ying-Shui Tan et al.
· arXiv.org · 18 citations
· ⚡1