Yuke Wang
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
MMA-SafetyBench: A Benchmark for Multimodal Agent Safety Evaluation
A preprint presents MMA-SafetyBench, 1,083 adversarial trajectories testing whether forged interface elements, such as fake system alerts, pull multimodal agents off their authorized task. The authors report high attack success across nine models.