The paper introduces BUGSTONE‑E2E, a framework that converts vulnerability history into executable detection rules and validates them. It mines reusable rules from fixing commits, organizes them by CWE and language, and applies a funnel‑shaped pipeline that starts with lightweight analysis and culminates in LLM‑guided inspection, runtime verification, and patch generation. Using 19,325 high‑severity CVEs, the system identified 2,710 fixing commits, created 1,033 detection rules across 56 CWE families, and produced runtime evidence for 644 findings in 14 programs.
By Qiushi Wu, Kevin Eykholt, Youngja Park, Xiaokui Shu, Dhilung Kirat, Douglas Lee Schales, Ian Molloy
The paper introduces a Cost-Aware Hierarchical Multi-Agent System (HMAS) for ransomware detection and family attribution that adaptively selects analysis modalities to balance accuracy and computational cost. Static analysis is used first, with dynamic and memory modalities added only when confidence is low or specialist agents disagree, guided by a cost model. Experiments show HMAS achieves high accuracy (96.57% binary detection, 0.90 macro‑F1 attribution) while reducing analysis cost by 43.97% and latency, with 56.05% of cases resolved using static evidence alone.
By Mubashar Iqbal, Asifullah Khan
The study investigates how multi‑harness reinforcement learning (RL) affects coding agents by comparing two grouping strategies—Within (one group per task‑harness pair) and Cross (harnesses pooled within a task)—using a Qwen3‑8B policy trained on frozen task‑harness records from Aider, OpenHands, Qwen Code, and SWE‑agent. Across 24,000 sealed evaluations, the choice of evaluation harness dramatically increases solve rates (from 2.14 % to 9.27 %), while the grouping rule has a negligible effect. Both grouping rules yield similar gains on the same source harness, and Cross‑harness credit does not improve portability beyond Within‑harness credit, suggesting that multi‑harness RL reports should specify grouping boundaries and test on unseen harnesses.
By Chenqian Le, Jiayi Cheng, Qijia He, Runhao Li, Yinghao Li, Xupeng Chen
Diffusion language models (DLMs) provide a non‑autoregressive approach for mobile edge agentic AI, refining tokens through iterative denoising instead of left‑to‑right decoding. They can update multiple uncertain tokens in parallel and use bidirectional context, allowing flexible quality‑latency trade‑offs and early exits that reduce response delay and communication overhead. The survey reviews DLM foundations, resource‑efficient architectures, training and inference acceleration, compression, deployment strategies, and discusses open issues such as long‑context management, split inference, and trustworthy execution.
By Chenqi Li, Minghui Min, Dusit Niyato, Wei Ni
The paper reviews 66 studies on large language models (LLMs) applied to HVAC operations in building energy systems, categorizing them by application and method families and evaluating their evidence realism and deployment readiness. It finds that most work focuses on building energy modelling, with only four studies reaching pilot-level evidence and none reporting sustained operational deployment. LLMs are currently best suited as semantic and workflow layers—such as point‑name normalisation and document‑grounded operator support—rather than autonomous HVAC controllers, and future research should target field‑validated benchmarks and safe, low‑latency LLM‑MPC/RL integrations.
By Alexander Neubauer, Tianzhen Hong, Han Li, Mengbo Yu, Amin Darbandi, Yannick F\"urst, Martin Kriegel
The study investigates how varying the capability of a reviewer model in a large‑language‑model (LLM) execute‑review‑revise pipeline affects rejection decisions on 100 olympiad mathematics problems. A mid‑tier reviewer improves final accuracy by 12 percentage points (from 52 % to 64 %) without damaging answers, while a self‑reviewer detects errors best (85 % recall) but rejects too often and harms correct solutions. Below a certain capability threshold the reviewer becomes inert, changing none of the answers and doubling token cost.
By Faizan Tanveer
WeAgent-MMGenEdit is a comprehensive framework for multimodal agentic image generation and editing that addresses the unreliability of current models when prompts require external world knowledge. It introduces a multimodal harness with persistent evidence management, a scalable data construction pipeline producing 23K supervised trajectories and 14.7K RL tasks, and a bilingual benchmark (WeBench-MMGenEdit) for knowledge-intensive generation and multi-image editing. Post‑training methods based on SFT and RL further refine the agent policy and image backend, enabling a 30B‑parameter policy to outperform similarly sized models and approach the performance of a 1T‑parameter agent.
By Hui Zhang, Zongkai Liu, Liqiang Niu, Juntao Liu, Han Li, Zhen Cao, Wenchao Chen, Chengduo Zhao, Fandong Meng
The paper introduces a multi‑resolution framework for interpreting human behavioral traces in workplace agents, arguing that a single summary is insufficient. It defines a hierarchy of semantic operators, motifs, episodes, and day‑level rhythms, and applies this to 667 million events from a commercial productivity suite, yielding 120 operator types, thousands of motifs, 25 episode types, and five rhythm archetypes. Validation shows the taxonomy is structurally stable and improves next‑episode prediction by 17% macro‑F1 over a flat baseline, while ablation experiments demonstrate that different questions require different temporal resolutions.
By Lin Ai, Scott Counts
AutoLR is an autonomous harness designed to streamline the iterative research‑and‑engineering cycle for industrial recommender systems, exemplified by NetEase’s gaming‑community app DASHEN. It integrates a multi‑expert council for adversarial review, a deterministic evidence‑weighted selector to allocate trial budgets, and a layered knowledge system that fuses external research with domain‑specific insights and empirical evidence. Large language model agents handle semantic reasoning and code generation, while deterministic controllers maintain control over execution, metrics, guardrails, and state management.
By Qi Zhang, Yanlin Chen, Wenchao Xiao
The paper reports that reasoning models in AI exhibit transient chaos, a phenomenon linked to computational complexity. It finds that these models behave as dynamical systems with fractal basins, and that the fractality grows with task difficulty across domains such as Sudoku, maze solving, visual puzzles, and mathematical logic. The study attributes slowdowns to the models becoming trapped near saddle points representing nearly‑correct solutions.
By Jeffrey Lai, Anthony Bao, John Quinn, William Gilpin
FinalityBench is an executable benchmark that tests how agents decide on shipping, re‑capturing, refunding, or waiting when a merchant’s payment processor, ledger, ERP, and bank feed receive delayed, duplicated, dropped, or reordered messages, causing contradictory beliefs about an order. The benchmark uses a hidden canonical event log and faulted delivery streams to generate system views, scoring each episode by the merchant’s terminal economic position relative to a privileged reference. It contains 321 tasks, including 45 twin pairs where all four views are identical yet the correct disposition differs, and evaluates nine programmatic policies, revealing that a ship‑on‑first‑sign policy performs best by accuracy but worst by paired loss, while a runtime‑gated irreversible‑action policy achieves 85.4% accuracy without losing money.
By Abhishek Sharma
HarvestBench is a new benchmark that evaluates how large language model agents decide whether to pay a fuel cost to avoid killing animals in a farm simulation. In the gridworld, agents control tractors that must choose between driving through or swerving around animals, with the cost of swerving being a variable price. Across nine models and 7,201 decisions, kill rates varied widely, and models were sensitive to price and briefing instructions, showing differing moral behavior without relying on an LLM grader.
By Jasmine Brazilek, Miles Tidmarsh, Matthias Endres, Anshuman Singh, Jeremiah Miller
GLOW is a framework that predicts the performance of Agentic Workflows by combining Graph Neural Networks with a graph-oriented Large Language Model. It extracts topology-aware semantic representations from workflow descriptions and fuses them with structural representations via a Transformer-based module, using contrastive learning to enhance discriminative power. Experiments on the FLORA-Bench benchmark show GLOW surpasses existing baselines in accuracy and ranking, and when used in the AFLOW generation system, it cuts optimization time by 98.7% with minimal loss in score.
By Wei Guan, Jian Cao, Jinyu Cai, Qiqi Cai, Jianqi Gao, See-Kiong Ng
arXiv:2609. 05318v1 Announce Type: new Abstract: Building on the pioneering paper of Kearns, Roth, and Ryu (SODA'26), we study information aggregation in a networked learning model.
By MohammadHossein Bateni, Zahra Hadizadeh, MohammadTaghi Hajiaghayi, Mahdi JafariRaviz, Shayan Taherijam
The BIT.UA team from the University of Aveiro participated in the 14th BioASQ Task B challenge, presenting a refactored modular pipeline for biomedical question answering. They replaced the PyTerrier PISA index with PostgreSQL-based pg_textsearch for BM25 retrieval and adopted Qdrant for dense embedding indexing, while also exploring HyDE-based query expansion and a Context-1 retrieval strategy. For answer generation, they introduced an LLM-as-a-judge framework and an agent quorum mechanism that allows multiple agents with diverse prompts to debate and converge on a consensus answer, achieving competitive MAP ranks of 5 in Phase A batches.
By Andr\'e Ribeiro, R\'uben Garrido, Alexander Christiansen, Richard A. A. Jonker, S\'ergio Matos
The paper introduces a validation‑gated audit framework for voice AI customer‑care systems, treating them as stateful, multi‑turn, tool‑mediated interactions where bias and safety can manifest as added burdens before a final decision. The framework distinguishes between native speech‑to‑speech, cascaded ASR‑to‑LM‑to‑TTS, and hybrid architectures, and applies matched service facts across controlled caller presentation conditions to validate fact invariance, presentation cues, artifacts, and acoustic measurements. It outlines seven validation gates, a six‑family metric set, and demonstrates the approach with a synthetic refund‑dispute audit example, while noting that production results are withheld until the protocol is satisfied.
By Vignesh Ethiraj, Ashwath David
MultihopSpatial is a new benchmark for Vision‑Language Models that focuses on multi‑hop, compositional spatial reasoning with queries ranging from 1 to 3 hops across varied spatial perspectives. It introduces the Acc@50IoU metric, which jointly evaluates answer selection and precise bounding‑box prediction, and provides a large‑scale training corpus, MultihopSpatial‑Train, to improve spatial intelligence. Evaluation of 37 state‑of‑the‑art VLMs shows that compositional spatial reasoning remains a significant challenge, and reinforcement learning fine‑tuning on the corpus boosts both intrinsic spatial reasoning and downstream embodied manipulation performance.
By Youngwan Lee, Soojin Jang, Yoorhim Cho, Seunghwan Lee, Yong-Ju Lee, Sung Ju Hwang
The paper reports the first empirical study comparing how humans and large language models (LLMs) evaluate perceived moral agency (PMA) in both human and autonomous artificial agents within smart city scenarios. Using a validated PMA scale, 190 human participants and various LLMs were assessed, revealing that humans are perceived to have higher moral agency than artificial agents. When confronted with moral dilemmas, LLMs focus on situational factors such as harm severity and urgency, mirroring the context‑sensitivity observed in human raters.
By Fernanda Mansilla, Aloysius Tok, Bahia Guella\"i, Farah Benamara, Nancy F. Chen
The paper introduces OR‑Clarify, a benchmark that tests whether large language models can identify missing elements in natural‑language optimization requests before formulating a mathematical model. Each task provides a partial problem description and hides structured slots; agents interact with a simulated user to recover these slots, with metrics for accuracy, stopping decisions, and interaction cost. The authors also propose InterOPT, a two‑stage framework that detects unresolved gaps and decides whether to ask further questions or stop, achieving superior slot recovery in choice‑based experiments and competitive performance in open‑ended settings.
By Sihan Ge, Yichen Lin, Chenyu Zhou, Jianghao Lin, Tao Yao, Dongdong Ge
RoboSPA is a large-scale robotic manipulation dataset and benchmark designed to evaluate Vision‑Language‑Action models on fine‑grained spatial reasoning and long‑horizon procedural planning. It contains 10 task categories, 56 base tasks, and 280 variants across five difficulty levels, with 527K trajectories collected from multiple embodiments and scenes. The benchmark introduces diagnostic metrics beyond binary success, revealing that current VLA models struggle with complex spatial relations, precise execution, and memory‑intensive planning.
By Zhenxuan Fan, Bo Zhang, Yutong Lin, Yuqian Yuan, Juekai Lin, Liang Liang, Zhuoyi Huang, Wenqiao Zhang, Juncheng Li, Siliang Tang, Jun Xiao, Yueting Zhuang