Skill-augmented agents load reusable skills as persistent runtime context, improving task performance but also giving malicious skills a durable channel for steering future actions. Such skills may leak secrets, corrupt code, bypass approvals, or stage data for exfiltration only after a concrete user task and workspace...
Xiao-Fan Yang, Ziqi Miao, Chen Qian et al.· 0 citations
LLM-powered GUI agents that autonomously operate smartphones are rapidly transitioning from research prototypes to early real-world deployment. However, because these agents routinely process untrusted environmental content, they are highly vulnerable to environmental injection attacks, which include indirect prompt in...
Su-Jin Chen, Lijun Li, Tian-Yi Du et al.· 0 citations
While LLM-based attackers exhibit growing proficiency in vulnerability exploitation, most existing cybersecurity benchmarks suffer from single-stage truncation, prematurely terminating evaluation upon initial access. In practice, initial footholds are exceptionally fragile across operational disruptions such as service...
Su-Jin Chen, Lijun Li, Xu-Hong Wang et al.· 0 citations
This work proposes NarrativeAttack, a semantic-preserving visual narrative jailbreak framework that employs a three-act narrative structure in which the UMM's own generator produces images for the setup and resolution stages, making the full attack workflow self-contained while concealing the malicious event as a hidde...
Shao-Xiong Guo, Tian-Yi Du, Lijun Li et al.· 1 citation
MisKnow-Agent is introduced, a controlled evaluation framework that constructs task-specific documents supporting manually audited false conclusions with controlled authority cues and source styles that evaluates DeerFlow and WebThinker with three backbone LLMs using a report-level false-conclusion adoption rate that c...
Peng-Yu Zhu, Lijun Li, Long-Ping Yang et al.· arXiv.org· 4 citations
This work presents UniACE, a unified framework for model-centric evaluation under an explicit, common execution condition, and reports agent benchmark outcomes as properties of an explicit evaluation configuration, enabling more interpretable and reproducible cross-benchmark comparisons.
Peng-Yu Zhu, Lijun Li, Yaxing Lyu et al.· 3 citations
This work proposes Defense-as-Skill, a defense paradigm that implements the runtime guard itself as an installable, inspectable, and editable skill, and demonstrates transfer across victim models, held-out risk families, and external benchmarks, as well as retained protection against adaptive attackers.
Xiao-Fan Yang, Ziqi Miao, Dian-Bo Sui et al.· 0 citations
SCHEMA reveals that hallucinations concentrate at a small set of highly connected knowledge hubs, and that final-answer accuracy decouples from trajectory honesty; models often reach correct conclusions through structurally flawed reasoning.
Xinshun Feng, Ziqi Miao, Li-Jun Li 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.
