AI Research Highlights

Dongxin Guo, Jikun Wu, Siu Ming Yiu

SafeAnchor reveals LLM safety is fragile and erodes cumulatively during domain adaptation. Practitioners must now actively preserve safety across updates—this is the first method to do so systematically in continual settings.

Zhonghao Zhan, Huichi Zhou, Zhenhao Li et al.

Introduces the 'Trust Gap' in agentic AI, revealing that tools can be weaponized to mislead agents—demanding new evaluation standards that test skepticism, not just competence, for real-world deployment safety.

Chenming Tang, Hsiu-Yuan Huang, Weijie Liu et al.

TRUSTEE trains tool-calling agents without labeled data or commercial models, using dynamic environment synthesis with only an 8B LLM—democratizing powerful agent training for any builder with minimal resources.

Zixuan Liu, Zhiyong Chen, Nan Xue et al.

WISV adapts speculative decoding verification to wireless conditions using semantic, not token-level, checks—dramatically improving edge-LLM latency and throughput in real-world mobile deployments.

Tao Zhang, Kaixian Qu, Zhibin Li et al.

DESPITE reveals that even highly accurate LLM planners can systematically fail safety-critical tasks, exposing a critical gap between planning accuracy and real-world safety—essential for deploying robots in human environments.

Jiaqi Li, Lvyang Zhang, Yang Zhao et al.

AIT Academy proposes the first principled curriculum for holistic agent development, addressing systemic gaps in current agent training—vital for builders aiming for general-purpose AI agents.

Wentao Zhang, Yan Zhuang, ZhuHang Zheng et al.

DEJA exposes stealthy RAG failures that mimic valid responses, forcing a paradigm shift in security evaluation—essential for deploying reliable RAG systems that must detect subtle, non-obvious degradation.

Christy Li, Sky CH-Wang, Andi Peng et al.

Human-Guided Harm Recovery introduces the first formal framework for correcting harmful agent actions post-execution, enabling safe, real-world deployment of AI agents with human-aligned recovery protocols.

Yujie Chen, Tailai Chen, Yifeng Gao et al.

Introduces delta attention halting that detects semantic fixing points to skip redundant token processing, enabling hardware-compatible efficiency gains in long-context LLMs without sacrificing accuracy—critical for deploying scalable inference.

Mina Gabriel, Pei Wang

Presents a neuro-symbolic pipeline translating natural language into Narsese, enabling interpretable, uncertainty-aware reasoning—vital for building trustworthy AI systems requiring explicit logic over LLM hallucinations.

Libo Sun, Peixiong He, Po-Wei Harn et al.

MoE-nD tailors KV cache compression per layer, boosting accuracy over uniform methods. Practitioners should care because it enables longer context inference with minimal memory overhead without retraining.

Ziyang Liu

Copy-as-Decode revolutionizes LLM editing by replacing full regeneration with grammar-constrained copy-gen operations, slashing latency and improving precision—critical for real-time code/text editing systems.

Tanjim Rahaman Fardin, S M Zunaid Alam, Mahadi Hasan Fahim et al.

MetaCloak-JPEG delivers JPEG-robust adversarial perturbations that block unauthorized DreamBooth deepfakes even after compression—essential for real-world privacy protection where images are routinely shared in degraded formats.

Eranga Bandara, Asanga Gunaratna, Ross Gore et al.

First on-device LLM deployment for psychiatric decision support that eliminates cloud egress, preserving patient privacy in high-risk settings. Enables real-time, compliant mental health AI without data leakage risks.

Zhenwen Liang, Yujun Zhou, Sidi Lu et al.

CUTS solves RL mode collapse in saturated reasoning by sampling from constrained top-K outputs, enabling continued learning even when models are already correct—vital for improving LLM reasoning robustness without manual data curation.