11 papers · 1 filter
RePolicy: Reinforcement Learning for Safety-Policy Invocation in Agent Safeguards
Houcheng Jiang, Boxuan Zhang, Qiyong Zhong +3
Safeguarding language model agents requires assessing complete execution trajectories under context-dependent safety policies. Existing policy-aware safeguards mainly rely on promp…
EasyOPD: An Easy-to-use On-Policy Distillation Framework for Large Language Models
Jie Sun, Mao Zheng, Mingyang Song +7
Conventional language-model distillation often relies on fixed teacher-generated data, which may not cover the states encountered by an evolving student policy. On-policy distillat…
Dynamic Spectral Denoising with Global-Context Attention for Multi-Behavior Recommendation
Miaomiao Cai, Yunshan Ma, Fangqi Zhu +5
Multi-behavior recommendation improves target-behavior prediction by exploiting heterogeneous auxiliary feedback (e.g., view, collect, and cart), yet its robustness is undermined b…
TRACE: Trajectory Risk-Aware Compression for Long-Horizon Agent Safety
Zhepei Hong, Lin Wang, Liting Li +5
Long-horizon LLM agents produce safety evidence across long trajectories, where sparse, delayed, and compositional risk signals often escape local moderation. Existing turn-level o…
SimCT: Recovering Lost Supervision for Cross-Tokenizer On-Policy Distillation
Jie Sun, Mao Zheng, Mingyang Song +6
On-policy distillation (OPD) is a standard tool for transferring teacher behavior to a smaller student, but it implicitly assumes that teacher and student predictions are comparabl…
Rubric-based On-policy Distillation
Junfeng Fang, Zhepei Hong, Mao Zheng +7
On-policy distillation (OPD) is a powerful paradigm for model alignment, yet its reliance on teacher logits restricts its application to white-box scenarios. We contend that struct…
SOD: Step-wise On-policy Distillation for Small Language Model Agents
Qiyong Zhong, Mao Zheng, Mingyang Song +5
Tool-integrated reasoning (TIR) is difficult to scale to small language models due to instability in long-horizon tool interactions and limited model capacity. While reinforcement…
A Systematic Security Evaluation of OpenClaw and Its Variants
Yuhang Wang, Haichang Gao, Zhenxing Niu +4
Tool-augmented AI agents substantially extend the practical capabilities of large language models, but they also introduce security risks that cannot be identified through model-on…
DRAFT: Task Decoupled Latent Reasoning for Agent Safety
Lin Wang, Junfeng Fang, Dan Zhang +3
The advent of tool-using LLM agents shifts safety monitoring from output moderation to auditing long, noisy interaction trajectories, where risk-critical evidence is sparse-making…
Enhancing Multi-Modal LLMs Reasoning via Difficulty-Aware Group Normalization
Jinghan Li, Junfeng Fang, Jinda Lu +5
Reinforcement Learning with Verifiable Rewards (RLVR) and Group Relative Policy Optimization (GRPO) have significantly advanced the reasoning capabilities of large language models.…
NExT-Guard: Training-Free Streaming Safeguard without Token-Level Labels
Junfeng Fang, Nachuan Chen, Houcheng Jiang +5
Large language models are increasingly deployed in streaming scenarios, rendering conventional post-hoc safeguards ineffective as they fail to interdict unsafe content in real-time…