#machine learning
May 2026
WMAttack: Automated Attack Search for Adversarial Evaluation of World-Model Agents
WMAttack, an automated attack-search framework for adversarial evaluation of world-model agents, is introduced and a theoretical explanation showing that proposal refinement improves finite-budget search when it shifts probability mass toward high-utility attacks is provided.
Zhixiang Guo, Si-Yuan Liang, Shi Fu et al.
· arXiv.org · 6 citations