13 papers
TRACES: Proactive Safety Auditing for Multi-Turn LLM Agents via Trajectory-State Modeling
Jiaqian Li, Yanshu Li, Boxuan Zhang +2
LLM agents increasingly operate through multi-turn tool use and environment interaction, where safety risks often emerge from intermediate steps long before they surface in the fin…
MemEye: A Visual-Centric Evaluation Framework for Multimodal Agent Memory
Minghao Guo, Qingyue Jiao, Zeru Shi +14
Long-term agent memory is increasingly multimodal, yet existing evaluations rarely test whether agents preserve the visual evidence needed for later reasoning. In prior work, many…
AgentForesight: Online Auditing for Early Failure Prediction in Multi-Agent Systems
Boxuan Zhang, Jianing Zhu, Zeru Shi +2
LLM-based multi-agent systems are increasingly deployed on long-horizon tasks, but a single decisive error is often accepted by downstream agents and cascades into trajectory-level…
A Single Layer to Explain Them All:Understanding Massive Activations in Large Language Models
Zeru Shi, Zhenting Wang, Fan Yang +2
We investigate the origins of massive activations in large language models (LLMs) and identify a specific layer named the \textbf{Massive Emergence Layer (ME Layer)}, that is consi…
Micro-Defects Expose Macro-Fakes: Detecting AI-Generated Images via Local Distributional Shifts
Boxuan Zhang, Jianing Zhu, Qifan Wang +2
Recent generative models can produce images that appear highly realistic, raising challenges in distinguishing real and AI-generated images. Yet existing detectors based on pre-tra…
From Rebound to Remedy: Understanding and Mitigating Reward Hacking via Representation Engineering
Rui Wu, Ruixiang Tang
Reinforcement learning for LLMs is vulnerable to reward hacking, where models exploit shortcuts to maximize reward without solving the intended task. We systematically study this p…