Bilevel data selection is a fundamental task with numerous applications in machine learning and statistics. A common strategy for tackling this problem is to relax the original bilevel combinatorial optimization into a continuous one. However, existing methods, including 0-1 relaxation and probabilistic relaxation, sti...
Shi-Zhan Cai, Qi-Xin Zhang, Si-Yuan Ma et al.· 0 citations
A unified taxonomy is established that organizes AI agent execution into three sequential stages: Reasoning, Acting, and Self-Evolving and systematically reviews the threats, defense mechanisms, and evaluation benchmarks associated with each component, providing a comprehensive understanding of security challenges thro...
Dai-Zong Liu, Tian-Yao Luo, Shu-Wei Huang et al.· AI Plus· 1 citation
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
To achieve effective, stealthy, and persistent control, TrojanWorld combines Decision-Reflective Induction to steer trigger-conditioned imagination toward attacker-specified actions using decision feedback, Clean Behavior Anchoring to preserve trigger-free predictive and behavioral fidelity, and Causal Propagation to s...
Wen-Kai Huang, Si-Yuan Liang, Gaolei Li et al.· 0 citations
This work experimentally and systematically analyzes the differences between clean and jailbreak samples in the cross-attention feature space, revealing for the first time a cumulative separation effect and a progressively increasing trend of linear separability between the two during the diffusion process.
Si-Yuan Liang, Yupeng Qiu, Junfeng Fang et al.· 0 citations
This work develops a four-part, intent-oriented taxonomy that organizes multi-turn jailbreaks by adversarial intent structure and finds that effectiveness is driven by how deliberately intent is organized across turns rather than by context length or query count.
Siyuan Li, Aodu Wulianghai, Zehao Liu et al.· 0 citations
This work studies the channel delivering search and page observations is a fragile security boundary and introduces Authority-Chain Hijack (ACH), an expert-refined strategy that turns isolated search-result and page-content manipulations into a coherent evidence chain across seemingly corroborating sources.
Xuebin Li, Han-Qing Zhao, Si-Yuan Liang et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
