Publications (7)
Direct Behavior Optimization: Unlocking the Potential of Lightweight LLMs
Hongming Yang, Shi Lin, Jun Shao +4
Lightweight Large Language Models (LwLLMs) are reduced-parameter, optimized models designed to run efficiently on consumer-grade hardware, offering significant advantages in resour…
LLMs can be Dangerous Reasoners: Analyzing-based Jailbreak Attack on Large Language Models
Shi Lin, Hongming Yang, Rongchang Li +4
The rapid development of Large Language Models (LLMs) has brought impressive advancements across various tasks. However, despite these achievements, LLMs still pose inherent safety…
Before Agents Speak: Pre-hoc Failure Risk Inference in Multi-Agent Systems
Shi Lin, Chenpei Wang, Peng Qian +4
The paper introduces HalluProp, a framework that predicts which agents in a large‑language‑model based multi‑agent system are likely to hallucinate and estimates the overall system…
A Survey of LLM-Driven AI Agent Communication: Protocols, Security Risks, and Defense Countermeasures
Dezhang Kong, Shi Lin, Zhenhua Xu +16
In recent years, Large-Language-Model-driven AI agents have exhibited unprecedented intelligence and adaptability. Nowadays, agents are undergoing a new round of evolution. They no…
Forecasting Trajectory-Level Safety Risks in Black-Box Multi-Turn Interactions
Shi Lin, Peng Qian, Dinghao Liu +5
The paper introduces Recast, a framework that predicts safety risks in multi‑turn interactions with large language models by forecasting how risks evolve over dialogue trajectories…
Web Fraud Attacks Against LLM-Driven Multi-Agent Systems
Dezhang Kong, Hujin Peng, Yilun Zhang +5
With the proliferation of LLM-driven multi-agent systems (MAS), the security of Web links has become a critical concern. Once MAS is induced to trust a malicious link, attackers ca…
Pushing the Limits of Safety: A Technical Report on the ATLAS Challenge 2025
Zonghao Ying, Siyang Wu, Run Hao +44
Multimodal Large Language Models (MLLMs) have enabled transformative advancements across diverse applications but remain susceptible to safety threats, especially jailbreak attacks…