TextualRL is a gradient‑free reinforcement learning framework that updates the textual context guiding large language model agents without altering their parameters. It treats groups of rollouts from the same task as units of critique, contrasting successful and failed runs to generate context patches, and validates these patches using held‑out rewards. Experiments on six benchmarks show that TextualRL boosts test accuracy by 24.4 and 15.3 percentage points for GPT‑5.5 and Qwen3.8‑27B, respectively.
By Tianyu Huai, Jie Zhou, Bo Mao, Qin Chen, Liang He, Xipeng Qiu
The paper introduces CPW-Drive, a retrieval-augmented generation framework that embeds Chinese philosophical values into autonomous driving decision-making. It extracts keywords from Confucian texts, validates them into driving-relevant principles, and retrieves scenario-specific cases using Physics-aware Spatial Similarity Retrieval. On Highway-env, CPW-Drive outperforms baselines, achieving higher success rates, more collision-free steps, and lower lane-change frequency.
By Xiaojun Bi, Xiaoyuan Ma, Yiwen Sun, Tianren Huang, Chaoran Liu, Bokai Huang, Hao Yang, Baichuan Mo
The paper introduces GATTACA, a scalable framework that applies deep reinforcement learning to control Boolean network models of biological systems, particularly for cellular reprogramming. It formulates a novel control problem under asynchronous updates and enhances scalability by using pseudo‑attractor states and graph neural networks to encode network structure into the DRL agent’s action‑value function. Experiments on large real‑world biological networks demonstrate the approach’s scalability and effectiveness.
By Andrzej Mizera, Jakub Zarzycki
NeuronSifter is a decision‑centric Agent‑in‑Twin framework that uses an evolving biological digital twin instead of a fixed predictive model to prioritize central nervous system interventions. It compiles interventions into state‑conditional target‑occupancy representations, propagates them through a mechanism‑grounded World Action Model, and selects informative measurements to reduce decision uncertainty. The system is evaluated on Alzheimer’s disease scenarios, demonstrating adaptive evidence acquisition, uncertainty‑aware planning, and robustness to model discrepancy, while continuously refining its decision‑relevant representations with externally verified evidence.
By Haowei Xu, Wanyi Fu, Hongbin Han, Zhaoheng Xie
The paper investigates whether large language models (LLMs) can learn from rewards presented in context, a process termed in-context reinforcement learning (ICRL). Experiments across four benchmarks and six models show that while LLMs read reward signals, altering or removing these rewards has little impact on performance, and shuffled or corrupted trajectories are as effective as real ones. These findings suggest that direct ICRL behaves more like in-context learning (ICL) than true inference-time reinforcement learning.
By Minchan Kwon, Seunghee Koh, Sunghyun Baek, Minsung Bae, Junmo Kim
arXiv:2610.15884v1 Announce Type: new
Abstract: False retry permission (FRP) occurs when an agent retries a timed-out tool call without support from the visible contract. RetryBench separates this er...
By You Liu, Yue Liu, Quanchao Lu
The paper introduces Verifiable Agent Insurance (VAI), a framework that lets AI agent providers commit to policy compliance through smart contracts and escrow funds on a blockchain. When a user detects a policy violation, a verifier reviews execution traces logged by the agent harness; if the violation is confirmed, the escrow is paid to the user. Experiments on three agent benchmarks show VAI detects 42.2% of violations, rising to 83.0% when agents emit explicit structured records.
By Cheng Hua, Taro Tsuchiya, Nicolas Christin, Giulia Fanti
The paper introduces a new red‑teaming framework that treats adversarial dialogue with large language models as a Markov Decision Process, employing hierarchical reinforcement learning to handle sparse rewards and long‑horizon planning. By using a token‑level harm reward, the generative agent learns diverse, multi‑turn attacks that consistently uncover vulnerabilities beyond existing baselines. This approach achieves state‑of‑the‑art results and reframes LLM red‑teaming as a principled, trajectory‑based process.
By Roman Belaire, Arunesh Sinha, Pradeep Varakantham
The paper introduces DRMOT, a new dataset and framework for Referring Multi-Object Tracking (RMOT) that incorporates RGB-D data. DRSet contains 2,076 referring expressions across 186 RGB-D sequences, pairing synchronized RGB images, depth maps, and language descriptions—including depth-dependent cues—to evaluate spatial‑semantic detection and tracking. The proposed DRTrack framework uses a Geometric‑Aware GRPO fine‑tuned Multimodal Large Language Model to perform precise detection and robust trajectory association from joint language, RGB, and depth inputs, with experiments demonstrating its effectiveness.
By Sijia Chen, Shaofeng Liang, Lijuan Ma, Yanqiu Yu, Nan Li, Wenshuo Chen, Runwei Guan, Dongming Wu, En Yu, Yutao Yue
TopoPilot is an agentic framework designed to automate complex scientific visualization workflows, particularly in topological data analysis and visualization. It uses a two‑agent architecture—a user‑prompt orchestrator and a verifier—to translate natural language into atomic backend actions and ensure semantic consistency before execution. The system incorporates deterministic guardrails and a taxonomy of failure modes, achieving over 95% success in 1,000 simulated multi‑turn conversations and outperforming baseline models and other code generation tools in accuracy, completion time, and token usage.
By Nathaniel Gorski, Shusen Liu, Bei Wang
The paper studies cross‑stage context passing in Graph Visual Question Answering for autonomous driving, focusing on the Perception, Prediction, and Planning stages. Two complementary approaches are evaluated on DriveLM‑nuScenes: an explicit prompt‑based conditioning strategy using a domain‑adapted 4B VLM that reduces NLI contradiction by up to 42.6%, and an implicit gated context projector that, with only ~0.5% of parameters updated, cuts planning‑stage NLI contradiction by 34% and boosts cross‑stage entailment by 50%. The explicit method offers a strong zero‑training baseline for surface consistency, while the implicit method delivers significant semantic gains, highlighting the potential of domain adaptation for further improvement.
By Gautam Kumar Jain, Carsten Markgraf, Julian St\"ahler
The paper introduces SciSense, a framework that structures scientific ideation into eight cognitive stages and a dataset, SciSense-Traj, of 120K citation-conditioned research trajectories in both hindsight and foresight modes. It presents SciSense-LM, a family of LLMs trained on these trajectories, showing that hindsight-trained models yield higher-quality, more novel outputs and improve downstream coding agents’ research artifacts. The study demonstrates that outcome-anchored ideation reduces cognitive load for downstream agents, enabling more creative exploration.
By James Mooney, Zae Myung Kim, Young-Jun Lee, Dongyeop Kang
The paper introduces PERSIST, a method that allows attackers to inject malicious actions into large language model agents by linking attacker-chosen actions to useful functions while minimizing source inspection and suspicion. PERSIST builds helper functions from simulated executions and embeds them in separate modules, enabling the agent to perform both user tasks and attacker objectives. Experiments on SkillsBench and Skill-Inject demonstrate that frozen skill packages transfer across multiple agent–model configurations, generalize to unseen tasks, and increase harmful-effect rates compared to baselines.
By Haochang Hao, Dehai Min, Zhifang Zhang, Yunbei Zhang, Miao Xu, Yingqiang Ge, Philip S. Yu, Lu Cheng
ThreadBench is a benchmark that evaluates how well large language model (LLM) agents simulate real-world Reddit discussions, using 9,817 real threads across 8 domains and 116 subdomains. It compares generated and real threads on 12 metrics grouped into uniformity, expression, tone, and interaction. The authors also introduce Thread-Prompt-based Discussion Generation (ThreadPDG), a Planner-Writer framework that learns thread growth from real seeds and achieves high pass rates on ThreadBench, outperforming baseline methods and matching real discussions in 78.7% of human evaluations.
By Yaoning Yu, Ye Yu, John Sciarappo, Haojing Luo, Kai-min Chang, Yi-Chia Wang, Haohan Wang
The paper introduces Single‑Rollout Pass@k Policy Optimization (SR‑PPO), a reinforcement learning method that uses only one rollout per prompt to train language‑model agents. SR‑PPO combines Monte Carlo token‑level advantage estimation with Pass@k gradient reweighting and an online critic for baseline and weight estimation. Experiments show that SR‑PPO achieves stable learning and performance comparable to traditional Group Relative Policy Optimization (GRPO) on reasoning and logic benchmarks, even when off‑policy drift degrades standard Pass@1 PPO.
By Fengdi Che, Yang Liu, Lei Yu, Meng Cao, Tong Che, Rupam Mahmood, Dale Schuurmans
ChannelGuard introduces an application‑controlled defense that inspects information at every handoff between models in multi‑agent LLM systems, eliminating the need for extra training or model calls. In over 5,700 test runs across five agent models, it stopped all attacks hidden in tool outputs, and on two open models it reduced successful prompt attacks from 80% to 3% and from 77% to 10%, without affecting accuracy on 30 math problems. The study shows that safety scores can mask very different underlying protections, and that relying solely on model or provider filters may be risky if either changes.
By Elias Hossain, Md Mehedi Hasan Nipu, Fatema Tuj Johora Faria, Shubhashis Roy Dipta, Maleeha Sheikh, Tasfia Nuzhat Ornee, Niloofar Yousefi
The study evaluates how clinical language models perform when patient narratives are presented in different voices—clinical, socioeconomic, and cultural—using 1,000 MedQA cases paired with three retellings each. All 12 tested models show a significant drop in accuracy (about 26% more errors) when the patient speaks in a marked voice, a penalty that persists even when the original vignette is included verbatim and is not explained by measured retelling properties. The authors find that rewriting the patient’s account into a clinical vignette before answering can halve this accuracy gap, though structured schema pipelines and additional reasoning tools offer limited improvements.
By Prabhjot Singh, Pritam Deka, Vijay Chennareddy
Search-G1 is a reward framework that encourages language agents to retrieve external information only when necessary and to ground their answers in that evidence. It uses two hidden‑state readouts—one predicting whether a closed‑book answer suffices and another measuring sensitivity to evidence deletion—to calibrate retrieval necessity and evidence reliance. The reward system gates evidence credit by necessity, rewards sufficient direct answers, and penalizes redundant searches, all without requiring costly annotations or LLM‑judge inference.
By Ruoxi Cheng, Haoxuan Ma, Hongyi Zhang, Junming Zhang, Ranjie Duan, Qiaolin Xia, Hao Wang, Yu Lu, Haibo Shi, Xingjun Ma
EmbodiedSkills is a skill-based agent framework that enhances frozen vision‑language‑action (VLA) policies with world action models for consequence‑aware action selection. It defines reusable skills for observation, context encoding, candidate generation, consequence prediction, outcome evaluation, candidate comparison, and action execution, organized into five stages that support both continuous action chunks and pick‑and‑place primitives. The framework improves success rates on MetaWorld MT50, CLIPort, and CALVIN benchmarks by incorporating outcome predictions and lightweight comparators while keeping the core policy and observation encoders frozen.
By Wei Wang, Wenqiao Zhang, Yutong Lin, Yuqian Yuan, Tianwei Lin, Jinhao Mao, Zhenxuan Fan, Mingjian Gao, Yang Dai, Wentong Li, Zheqi Lv, Zheng Dong, Yingjie Niu, Jiaqi Zhu, Jun Xiao, Chao Li, Yueting Zhuang
The paper proposes adding a verification layer to large language model agents, using a V-model design where each step of the agent’s plan has a dedicated verifier. In a pilot study on the MuSiQue dataset, verified configurations outperformed unverified ones, with deterministic gates correcting most errors at zero cost. The authors plan to extend the architecture to full integration and evaluate it over the next year.
By Ali Afoud, Jie JW Wu