13 papers
Feature-Aware Token Attack for Compression-Triggered Stealthy Failures in Large Vision-Language Models
Shilinlu Yan, Bowen Chen, Yuechen Zhang +3
Visual-token compression improves the efficiency of large vision-language models, but can expose failures that full-token evaluation misses. We study adversarial images that preser…
When Reasoning Goes Astray: Attention Dynamics of Uncontrolled Reasoning
Yuanhe Zhang, Ziwei Wang, Jie Ren +6
Large reasoning models (LRMs) improve performance on complex tasks through extended reasoning, yet the same process can degenerate into redundant verification and persistent genera…
Uncovering Uncontrolled Repetition through Residual Stream Dynamics
Yuanhe Zhang, Xinyao Zhou, Haoran Gao +5
Uncontrolled repetition can prolong autoregressive generation in large language models (LLMs) and enable resource consumption attacks. Prior analyses of repetitive generation have…
RESCUE: Repairing Language Model Errors to Sparse Circuits via Reinforcement Learning
Chuanpu Liu, Miao Yu, Yikai Cai +5
Large language models (LLMs) exhibit strong general capabilities that mechanistic interpretability has attributed to sparse computational circuits. However, existing circuit studie…
From Inaudible Inputs to Model Failures: Low-Frequency Safety Risks in LALMs
Yuanhe Zhang, Weiliu Wang, Jie Ren +7
Large audio-language models (LALMs) have demonstrated strong capabilities in understanding diverse audio inputs. This diversity includes low-frequency signals that are inaudible to…
LARFT: Closing the Cognition-Action Gap for Length Instruction Following in Large Language Models
Wei Zhang, Lintong Du, Yuanhe Zhang +4
Despite the strong performance of Large Language Models (LLMs) on complex instruction-following tasks, precise control of output length remains a persistent challenge. Existing met…
From Helpfulness to Toxic Proactivity: Diagnosing Behavioral Misalignment in LLM Agents
Xinyue Wang, Yuanhe Zhang, Zhengshuo Gong +6
The enhanced capabilities of LLM-based agents come with an emergency for model planning and tool-use abilities. Attributing to helpful-harmless trade-off from LLM alignment, agents…
SEE: Signal Embedding Energy for Quantifying Noise Interference in Large Audio Language Models
Yuanhe Zhang, Jiayu Tian, Yibo Zhang +5
Large Audio Language Models (LALMs) have been widely applied in real-time scenarios, such as in-car assistants and online meeting comprehension. In practice, audio inputs are often…
HearSay Benchmark: Do Audio LLMs Leak What They Hear?
Jin Wang, Liang Lin, Kaiwen Luo +8
While Audio Large Language Models (ALLMs) have achieved remarkable progress in understanding and generation, their potential privacy implications remain largely unexplored. This pa…
RSA-Bench: Benchmarking Audio Large Models in Real-World Acoustic Scenarios
Yibo Zhang, Liang Lin, Kaiwen Luo +8
While Audio Large Models (ALMs) have achieved remarkable proficiency, their robustness remains brittle in real-world deployment. Existing evaluations largely rely on synthetic Gaus…
ChronosAudio: A Comprehensive Long-Audio Benchmark for Evaluating Audio-Large Language Models
Kaiwen Luo, Liang Lin, Yibo Zhang +8
Although Audio Large Language Models (ALLMs) have witnessed substantial advancements, their long audio understanding capabilities remain unexplored. A plethora of benchmarks have b…
LeechHijack: Covert Computational Resource Exploitation in Intelligent Agent Systems
Yuanhe Zhang, Weiliu Wang, Zhenhong Zhou +5
Large Language Model (LLM)-based agents have demonstrated remarkable capabilities in reasoning, planning, and tool usage. The recently proposed Model Context Protocol (MCP) has eme…
Resource Consumption Red-Teaming for Large Vision-Language Models
Haoran Gao, Yuanhe Zhang, Zhenhong Zhou +7
Resource Consumption Attacks (RCAs) have emerged as a significant threat to the deployment of Large Language Models (LLMs). With the integration of vision modalities, additional at…