Dongrui Liu

Mingxuan Zhang, Jiahui Han, Dadi Guo et al.

LLM-based agents are rapidly advancing, autonomously invoking external tools to complete multi-step tasks for users. However, agents often acquire more sensitive information than the task requires. Existing privacy benchmarks audit what the agent's response or outgoing actions disclose, but overlook the acquisition stage where data first enters the agent's context. The over-acquired information is then one careless action or one attack away from an outright leak. To assess its prevalence, we introduce \emph{PrivacyPeek}, a benchmark for evaluating acquisition-stage privacy leakage of LLM-based agents, with $1{,}182$ cases across $7$ acquisition behaviours and $16$ application domains. Specifically, \emph{Acquisition Inspection} examines the agent's tool-call trajectory, both the tools it invokes and the data it receives, to detect when it acquires sensitive information beyond the task scope. \emph{Probe Elicitation} then issues a follow-up probe and measures how readily an attacker could elicit sensitive information the agent acquired but did not disclose. Our experiments on 10 LLM-based agents across 4 model families show that the unnecessary acquisition of sensitive information is widespread. In addition, we observe a correlation between the task-completion capability and acquisition-stage leakage. Prompt-level defences reduce only a small fraction of acquisition-stage leakage, leaving the majority unmitigated. These results make auditing acquisition-stage privacy both urgent and necessary. Our dataset and code are available at https://github.com/Xuan269/PrivacyPeek-Resource.

Xinhao Song, Su Su, Sirui Song et al.

Multimodal agents are increasingly expected to operate interfaces on behalf of users, raising a central deployment question: can they truly substitute for humans in workflows that services deliberately protect against automation? CAPTCHA verification makes this question concrete. It is not merely a visual puzzle, but a human-verification boundary placed before account creation, content access, form submission, and other protected actions. We introduce \textbf{Humanity's Last Line of Verification (HLL)}, a controlled benchmark that uses interactive CAPTCHA verification to evaluate whether agents can cross this boundary through grounded, human-like interaction rather than recognition alone. HLL covers diverse CAPTCHA interactions and exposes agents to controlled realism stressors, including cluttered webpages, harder task variants, and trace-conditioned validation of the solving process. We evaluate eight frontier multimodal agents in a closed-loop GUI environment. The results show that current agents remain brittle at this human-substitution boundary: performance varies sharply across verification types, degrades under realistic interface conditions, and drops further when correct answers must be supported by valid action traces. By exposing gaps in localization, action calibration, state tracking, and process consistency, HLL provides a concrete testbed for measuring how close multimodal agents are to acting as human substitutes in protected real-world workflows. Our code is available at https://github.com/XinhaoS0101/HLL

Tianyi Zhou, Dongrui Liu, Leitao Yuan et al.

LLM agents are increasingly expected not only to complete isolated tasks, but also to carry bounded representations of human expertise, judgment, and interaction style. Building such person-grounded agents remains difficult because actionable knowledge associated with a person or role is usually embedded in heterogeneous traces rather than written as clean instructions. Existing memory and persona systems capture fragments of this evidence, while skill frameworks provide portable packaging formats; however, there is no end-to-end workflow for distilling these traces into inspectable, correctable, and agent-usable skills. We present an automated trace-to-skill distillation system for generating person-grounded AI skills via expert knowledge distillation. Given materials from a target person or role, COLLEAGUE.SKILL produces a versioned skill package with two coordinated tracks: a capability track for practices, mental models, and decision heuristics, and a bounded behavior track for communication style, interaction rules, and correction history. The package can be inspected, invoked, updated through natural-language feedback, rolled back, installed across agent hosts, and optionally prepared for controlled distribution. We describe the artifact contract, generation workflow, correction lifecycle, deployment surface, and domain presets implemented in the open-source system. At the time of writing, the public repository has approximately 18.5k GitHub stars; the gallery lists 215 skills from 165 contributors and more than 100k cumulative stars across listed skill cards. The system illustrates how person-grounded skills can be represented as portable, correctable packages rather than opaque prompts or hidden memories.

Yongxiang Li, Moxin Li, Zhixin Ma et al.

Large Language Model (LLM) agents remain vulnerable to safety threats from the external environment, where attackers inject adversarial content into external observations such as tool-returned data, webpages, or MCP context, causing harmful agentic behaviors such as unsafe actions or incorrect outputs. Existing studies typically focus on single-interaction attacks, where the agent observes adversarial content and immediately exhibits harmful behavior within one user request. However, we show that adversarial content can also persist across interactions served by the same agent, making such threats harder to detect and mitigate. Specifically, adversarial content may persist in the agent state, remain dormant across interactions, and later be activated by a benign user query. We formalize this type of safety threat as Sleeper Attack. To evaluate it, we construct a benchmark with 1,896 instances covering six real-world harmful outcomes, three attack strategies, and three agent state targets: session context, memory, and reusable skills. Experiments on seven strong open-source and closed-source LLMs show that state-of-the-art LLM agents remain vulnerable to Sleeper Attack, even when they achieve low attack success rates under a single-interaction baseline. Our code and data are available at https://anonymous.4open.science/r/skdvnfu23ihr9wdscnksf1asdffsaef.

Huilin Zhou, Hao Zhang, Huiqi Deng et al.

This paper explains the generalization power of a deep neural network (DNN) from the perspective of interactions. Although there is no universally accepted definition of the concepts encoded by a DNN, the sparsity of interactions in a DNN has been proved, i.e., the output score of a DNN can be well explained by a small number of interactions between input variables. In this way, to some extent, we can consider such interactions as interactive concepts encoded by the DNN. Therefore, in this paper, we derive an analytic explanation of inconsistency of concepts of different complexities. This may shed new lights on using the generalization power of concepts to explain the generalization power of the entire DNN. Besides, we discover that the DNN with stronger generalization power usually learns simple concepts more quickly and encodes fewer complex concepts. We also discover the detouring dynamics of learning complex concepts, which explains both the high learning difficulty and the low generalization power of complex concepts. The code will be released when the paper is accepted.

Leitao Yuan, Qinghua Mao, Daizong Liu et al.

Multimodal large language models (MLLMs) remain vulnerable to transfer-based targeted attacks, where perturbations optimized on open-source surrogate encoders can generalize to closed-source MLLMs. A key challenge for improving adversarial transferability is to effectively capture the intrinsic visual focus shared across different models, such that perturbations align with transferable semantic cues rather than surrogate-specific behaviors. However, existing methods suffer from spatial-domain feature redundancy and surrogate-specific gradient signals, thereby hindering cross-model transferability. In this paper, we propose FRA-Attack, which addresses both challenges from a unified frequency-domain regularization perspective. For feature alignment, a high-pass DCT objective on patch features suppresses redundant global structures and concentrates the loss on the high-frequency band that carries the MLLMs' intrinsic visual focus. For gradient optimization, we introduce Frequency-domain Gradient Regularization (FGR), a \textit{model-agnostic} low-pass regularizer that modulates the surrogate gradient using only the geometric frequency coordinate, \textit{i.e.}, no surrogate-derived statistic is involved, so that FGR is model-agnostic by construction, removing surrogate-specific high-frequency artifacts while preserving transferable low-frequency directions. Together, the two components form a unified frequency-domain treatment of transferability. Extensive experiments on $15$ flagship MLLMs across $7$ vendors show that FRA-Attack achieves superior cross-model transferability, particularly with state-of-the-art performance on GPT-5.4, Claude-Opus-4.6 and Gemini-3-flash.

Kaiwen Luo, Zhenhong Zhou, Leo Wang et al.

The foundational capabilities established by Large Language Models (LLMs) have paved the way for Multimodal Large Language Models (MLLMs), within which Large Audio Language Models (LALMs) are essential for realizing universal auditory intelligence. Despite their remarkable performance, the escalation of LALMs' capabilities has significantly outpaced the development of systemic frameworks to ensure their trustworthiness. This survey provides a comprehensive investigation into the endogenous mechanisms of LALMs, detailing the architectural innovations and alignment algorithms that facilitate emergent reasoning. Specifically, we analyze how the transition to unified end-to-end frameworks and the integration of continuous acoustic signals inherently expand the attack surface. To rigorously evaluate the risks within these paradigms, we establish a comprehensive taxonomy of trustworthiness, categorizing critical vulnerabilities such as cross-modal jailbreaking, latent acoustic backdoors, and biometric privacy leakage. We review the state-of-the-art through six analytical pillars: hallucination, robustness, safety, privacy, fairness, and authentication. The profound imbalance between a mature offensive landscape and underdeveloped defenses further validates the critical trustworthiness gaps and multidimensional risks facing audio-centric intelligence. Finally, we propose a strategic roadmap advocating for "Defense-in-Depth" architectures, causal auditory world modeling, and intrinsic representation engineering to bridge the gap between empirical performance and intrinsically trustworthy audio intelligence. Our project has been uploaded to GitHub https://github.com/Kwwwww74/Awesome-Trustworthy-AudioLLMs.

Dongrui Liu, Yu Li, Zhonghao Yang et al.

Modern open-world agents such as OpenClaw exhibit powerful cross-environment execution capabilities yet introduce broad new safety risk sources. Meanwhile, advanced frontier AI models drastically lower attack barriers, rendering current agent alignment frameworks inadequate for real-world deployment. To tackle these emerging threats, we propose a lightweight and scalable agent safety alignment framework. Specifically, we update the agent safety taxonomy to accommodate emergent risks from Codex and OpenClaw execution scenarios. We further build a taxonomy-guided data engine with influence-function purification to train lightweight AgentDoG 1.5 variants (0.8B, 2B, 4B, and 8B parameters) using only around 1k samples, achieving comparable performance with leading closed-source models (e.g., GPT-5.4). Based on AgentDoG 1.5, we construct a highly efficient agentic safety SFT and RL training environment, which reduces deployment overhead in Docker-level environments by two orders of magnitude. Finally, we deploy AgentDoG 1.5 as a training-free online guardrail for real-time safety moderation. Extensive experimental results indicate that AgentDoG 1.5 achieves state-of-the-art performance in diverse and complex interactive agentic scenarios. All models and datasets are openly released.

Xiaoya Lu, Yijin Zhou, Zeren Chen et al.

Vision-Language Models (VLMs) empower embodied agents to execute complex instructions, yet they remain vulnerable to contextual safety risks where benign commands become hazardous due to subtle environmental states. Existing safeguards often prove inadequate. Rule-based methods lack scalability in object-dense scenes, whereas model-based approaches relying on prompt engineering suffer from unfocused perception, resulting in missed risks or hallucinations. To address this, we propose an architecture-agnostic safeguard featuring Context-Guided Chain-of-Thought (CG-CoT). This mechanism decomposes risk assessment into active perception that sequentially anchors attention to interaction targets and relevant spatial neighborhoods, followed by semantic judgment based on this visual evidence. We support this approach with a curated grounding dataset and a two-stage training strategy utilizing Reinforcement Fine-Tuning (RFT) with process rewards to enforce precise intermediate grounding. Experiments demonstrate that our model HomeGuard significantly enhances safety, improving risk match rates by over 30% compared to base models while reducing oversafety. Beyond hazard detection, the generated visual anchors serve as actionable spatial constraints for downstream planners, facilitating explicit collision avoidance and safety trajectory generation. Code and data are released under https://github.com/AI45Lab/HomeGuard

Lewen Yan, Jilin Mei, Tianyi Zhou et al.

LLM-based trading agents are increasingly deployed in real-world financial markets to perform autonomous analysis and execution. However, their reliability and robustness under adversarial or faulty conditions remain largely unexamined, despite operating in high-risk, irreversible financial environments. We propose TradeTrap, a unified evaluation framework for systematically stress-testing both adaptive and procedural autonomous trading agents. TradeTrap targets four core components of autonomous trading agents: market intelligence, strategy formulation, portfolio and ledger handling, and trade execution, and evaluates their robustness under controlled system-level perturbations. All evaluations are conducted in a closed-loop historical backtesting setting on real US equity market data with identical initial conditions, enabling fair and reproducible comparisons across agents and attacks. Extensive experiments show that small perturbations at a single component can propagate through the agent decision loop and induce extreme concentration, runaway exposure, and large portfolio drawdowns across both agent types, demonstrating that current autonomous trading agents can be systematically misled at the system level. Our code is available at https://github.com/Yanlewen/TradeTrap.

Bo Zhang, Jiaxuan Guo, Lijun Li et al.

As the development of Large Models (LMs) progresses rapidly, their safety is also a priority. In current Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) safety workflow, evaluation, diagnosis, and alignment are often handled by separate tools. Specifically, safety evaluation can only locate external behavioral risks but cannot figure out internal root causes. Meanwhile, safety diagnosis often drifts from concrete risk scenarios and remains at the explainable level. In this way, safety alignment lack dedicated explanations of changes in internal mechanisms, potentially degrading general capabilities. To systematically address these issues, we propose an open-source project, namely DeepSight, to practice a new safety evaluation-diagnosis integrated paradigm. DeepSight is low-cost, reproducible, efficient, and highly scalable large-scale model safety evaluation project consisting of a evaluation toolkit DeepSafe and a diagnosis toolkit DeepScan. By unifying task and data protocols, we build a connection between the two stages and transform safety evaluation from black-box to white-box insight. Besides, DeepSight is the first open source toolkit that support the frontier AI risk evaluation and joint safety evaluation and diagnosis.

Jie Zhang, Dongrui Liu, Chen Qian et al.

Personality psychologists have analyzed the relationship between personality and safety behaviors in human society. Although Large Language Models (LLMs) demonstrate personality traits, the relationship between personality traits and safety abilities in LLMs still remains a mystery. In this paper, we discover that LLMs' personality traits are closely related to their safety abilities, i.e., toxicity, privacy, and fairness, based on the reliable MBTI-M scale. Meanwhile, the safety alignment generally increases various LLMs' Extraversion, Sensing, and Judging traits. According to such findings, we can edit LLMs' personality traits and improve their safety performance, e.g., inducing personality from ISTJ to ISTP resulted in a relative improvement of approximately 43% and 10% in privacy and fairness performance, respectively. Additionally, we find that LLMs with different personality traits are differentially susceptible to jailbreak. This study pioneers the investigation of LLM safety from a personality perspective, providing new insights into LLM safety enhancement.

Dadi Guo, Yuejin Xie, Qingyu Liu et al.

As large language models (LLMs) advance their mathematical capabilities toward the IMO level, the scarcity of challenging, high-quality problems for training and evaluation has become a significant bottleneck. Simultaneously, recent code agents have demonstrated sophisticated skills in agentic coding and reasoning, suggesting that code execution can serve as a scalable environment for mathematical experimentation. In this paper, we investigate the potential of code agents to autonomously evolve existing math problems into more complex variations. We introduce a multi-agent framework designed to perform problem evolution while validating the solvability and increased difficulty of the generated problems. Our experiments demonstrate that, given sufficient test-time exploration, code agents can synthesize new, solvable problems that are structurally distinct from and more challenging than the originals. This work provides empirical evidence that code-driven agents can serve as a viable mechanism for synthesizing high-difficulty mathematical reasoning problems within scalable computational environments. Our data is available at https://github.com/TarferSoul/Code2Math.

Chen Xu, Yicheng Hu, Ruizi Wang et al.

Multi-agent systems (MAS) have emerged as a promising paradigm for solving complex tasks. Recent work has explored self-evolving MAS that automatically optimize agent capabilities or communication topologies. However, existing methods either learn a topology that remains fixed at inference time or adapt only the topology or capability during inference. We empirically and theoretically show that effective test-time evolution requires jointly adapting both axes, but on different time scales: capabilities should update rapidly to handle emerging subtasks, while the topology should evolve more slowly to preserve coordination stability. We then introduce TacoMAS, a test-time co-evolution framework for dynamic MAS. TacoMAS formulates MAS inference as a task of online graph adaptation, where nodes represent agents with role-specific capabilities and edges define their communication topology. During inference, a fast capability loop updates agent expertise using trajectory-level feedback, while a slow meta-LLM-driven topology loop performs agents' birth-death operations on MAS, including edge edit, agent addition, and agent removal. We further show that this fast-slow design drives MAS evolution toward a task-conditioned stable equilibrium. Experiments on four benchmarks demonstrate that TacoMAS outperforms nearly 20 multi-agent baselines, achieving an average improvement of 13.3% over the strongest baseline. The codes are released at https://github.com/chenxu2-gif/TacoMAS-MultiAgent.

Zhida He, Xiaoyu Wen, Han Qi et al.

Deploying LLMs in multi-turn dialogues facilitates jailbreak attacks that distribute harmful intent across seemingly benign turns. Recent training-based multi-turn jailbreak methods learn long-horizon attack strategies from interaction feedback, but often rely on coarse trajectory-level outcome signals that broadcast uniformly to every turn. However, we find that turn-level contributions in multi-turn jailbreaking are non-uniform, phase-dependent, and target-specific. Such coarse outcome supervision induces a credit assignment problem, leading to over-rewarding redundant turns in successful trajectories and under-crediting useful intermediate turns in failed ones. To address this, we propose TRACE, a turn-aware credit assignment framework for reinforcement learning (RL)-based multi-turn jailbreaking. For successful trajectories, TRACE estimates turn-level contributions via leave-one-turn-out semantic masking; for failed ones, TRACE assigns penalties based on prompt harmfulness and semantic relevance, with an additional local refusal-aware penalty. Furthermore, we reuse the attack-side credit signal for multi-turn defense alignment. Extensive experiments on open-source and closed-source targets show that TRACE achieves strong overall performance in effectiveness, transferability, and efficiency, yielding about a 25% relative improvement in attack success rate over the strongest RL baseline while also improving the safety-utility balance when reused for defense alignment.

Yulang Chen, Haoxuan Peng, Jinyan Liu et al.

Large Language Model-based Multi-Agent Systems (MAS) have demonstrated remarkable capabilities in complex tasks. However, manually designing optimal communication topologies is labor-intensive, while automated expansion methods often result in bloated structures with redundant agents, leading to excessive token consumption. To address this problem, we introduce \textbf{AgentSlimming}, a plug-and-play compression framework for graph-structured multi-agent workflows. Motivated by pruning and quantization in neural networks, AgentSlimming compresses workflows by first estimating the importance score of each agent with a hybrid mechanism, and then removes redundant agents or replaces them with low-cost ones, where each operation is validated using a baseline-anchored acceptance rule to prevent performance collapse. Experiments show that AgentSlimming reduces average token cost by up to 78.9\% with negligible performance degradation, and sometimes even improves accuracy, achieving a strong Pareto-optimal trade-off between cost and quality. \textit{Our code is publicly available at https://github.com/CitrusYL/AgentSlimming

Xuhao Hu, Dongrui Liu, Hao Li et al.

Safety concerns of Multimodal large language models (MLLMs) have gradually become an important problem in various applications. Surprisingly, previous works indicate a counterintuitive phenomenon that using textual unlearning to align MLLMs achieves comparable safety performances with MLLMs aligned with image text pairs. To explain such a phenomenon, we discover a Visual Safety Information Leakage (VSIL) problem in existing multimodal safety benchmarks, i.e., the potentially risky content in the image has been revealed in the textual query. Thus, MLLMs can easily refuse these sensitive image-text pairs according to textual queries only, leading to unreliable cross-modality safety evaluation of MLLMs. We also conduct a further comparison experiment between textual alignment and multimodal alignment to highlight this drawback. To this end, we construct multimodal Visual Leakless Safety Bench (VLSBench) with 2.2k image-text pairs through an automated data pipeline. Experimental results indicate that VLSBench poses a significant challenge to both open-source and close-source MLLMs, e.g., LLaVA, Qwen2-VL and GPT-4o. Besides, we empirically compare textual and multimodal alignment methods on VLSBench and find that textual alignment is effective enough for multimodal safety scenarios with VSIL, while multimodal alignment is preferable for safety scenarios without VSIL. Code and data are released under https://github.com/AI45Lab/VLSBench

Qibing Ren, Hao Li, Dongrui Liu et al.

Safety concerns in large language models (LLMs) have gained significant attention due to their exposure to potentially harmful data during pre-training. In this paper, we identify a new safety vulnerability in LLMs: their susceptibility to \textit{natural distribution shifts} between attack prompts and original toxic prompts, where seemingly benign prompts, semantically related to harmful content, can bypass safety mechanisms. To explore this issue, we introduce a novel attack method, \textit{ActorBreaker}, which identifies actors related to toxic prompts within pre-training distribution to craft multi-turn prompts that gradually lead LLMs to reveal unsafe content. ActorBreaker is grounded in Latour's actor-network theory, encompassing both human and non-human actors to capture a broader range of vulnerabilities. Our experimental results demonstrate that ActorBreaker outperforms existing attack methods in terms of diversity, effectiveness, and efficiency across aligned LLMs. To address this vulnerability, we propose expanding safety training to cover a broader semantic space of toxic content. We thus construct a multi-turn safety dataset using ActorBreaker. Fine-tuning models on our dataset shows significant improvements in robustness, though with some trade-offs in utility. Code is available at https://github.com/AI45Lab/ActorAttack.

Dongrui Liu, Qihan Ren, Chen Qian et al.

The rise of AI agents introduces complex safety and security challenges arising from autonomous tool use and environmental interactions. Current guardrail models lack agentic risk awareness and transparency in risk diagnosis. To introduce an agentic guardrail that covers complex and numerous risky behaviors, we first propose a unified three-dimensional taxonomy that orthogonally categorizes agentic risks by their source (where), failure mode (how), and consequence (what). Guided by this structured and hierarchical taxonomy, we introduce a new fine-grained agentic safety benchmark (ATBench) and a Diagnostic Guardrail framework for agent safety and security (AgentDoG). AgentDoG provides fine-grained and contextual monitoring across agent trajectories. More Crucially, AgentDoG can diagnose the root causes of unsafe actions and seemingly safe but unreasonable actions, offering provenance and transparency beyond binary labels to facilitate effective agent alignment. AgentDoG variants are available in three sizes (4B, 7B, and 8B parameters) across Qwen and Llama model families. Extensive experimental results demonstrate that AgentDoG achieves state-of-the-art performance in agentic safety moderation in diverse and complex interactive scenarios. All models and datasets are openly released.

Jie Zhang, Dongrui Liu, Chen Qian et al.

Protecting the intellectual property of open-source Large Language Models (LLMs) is very important, because training LLMs costs extensive computational resources and data. Therefore, model owners and third parties need to identify whether a suspect model is a subsequent development of the victim model. To this end, we propose a training-free REEF to identify the relationship between the suspect and victim models from the perspective of LLMs' feature representations. Specifically, REEF computes and compares the centered kernel alignment similarity between the representations of a suspect model and a victim model on the same samples. This training-free REEF does not impair the model's general capabilities and is robust to sequential fine-tuning, pruning, model merging, and permutations. In this way, REEF provides a simple and effective way for third parties and models' owners to protect LLMs' intellectual property together. The code is available at https://github.com/tmylla/REEF.

Tianyu Chen, Chujia Hu, Ge Gao et al.

Computer-use agents (CUAs) that interact with real computer systems can perform automated tasks but face critical safety risks. Ambiguous instructions may trigger harmful actions, and adversarial users can manipulate tool execution to achieve malicious goals. Existing benchmarks mostly focus on short-horizon or GUI-based tasks, evaluating on execution-time errors but overlooking the ability to anticipate planning-time risks. To fill this gap, we present LPS-Bench, a benchmark that evaluates the planning-time safety awareness of MCP-based CUAs under long-horizon tasks, covering both benign and adversarial interactions across 65 scenarios of 7 task domains and 9 risk types. We introduce a multi-agent automated pipeline for scalable data generation and adopt an LLM-as-a-judge evaluation protocol to assess safety awareness through the planning trajectory. Experiments reveal substantial deficiencies in existing CUAs' ability to maintain safe behavior. We further analyze the risks and propose mitigation strategies to improve long-horizon planning safety in MCP-based CUA systems. We open-source our code at https://github.com/tychenn/LPS-Bench.

Qihan Ren, Junpeng Zhang, Yang Xu et al.

This study proves the two-phase dynamics of a deep neural network (DNN) learning interactions. Despite the long disappointing view of the faithfulness of post-hoc explanation of a DNN, a series of theorems have been proven in recent years to show that for a given input sample, a small set of interactions between input variables can be considered as primitive inference patterns that faithfully represent a DNN's detailed inference logic on that sample. Particularly, Zhang et al. have observed that various DNNs all learn interactions of different complexities in two distinct phases, and this two-phase dynamics well explains how a DNN changes from under-fitting to over-fitting. Therefore, in this study, we mathematically prove the two-phase dynamics of interactions, providing a theoretical mechanism for how the generalization power of a DNN changes during the training process. Experiments show that our theory well predicts the real dynamics of interactions on different DNNs trained for various tasks.

Zhihua Wei, Qiang Li, Jian Ruan et al.

Large vision-language models (VLMs) often exhibit weakened safety alignment with the integration of the visual modality. Even when text prompts contain explicit harmful intent, adding an image can substantially increase jailbreak success rates. In this paper, we observe that VLMs can clearly distinguish benign inputs from harmful ones in their representation space. Moreover, even among harmful inputs, jailbreak samples form a distinct internal state that is separable from refusal samples. These observations suggest that jailbreaks do not arise from a failure to recognize harmful intent. Instead, the visual modality shifts representations toward a specific jailbreak state, thereby leading to a failure to trigger refusal. To quantify this transition, we identify a jailbreak direction and define the jailbreak-related shift as the component of the image-induced representation shift along this direction. Our analysis shows that the jailbreak-related shift reliably characterizes jailbreak behavior, providing a unified explanation for diverse jailbreak scenarios. Finally, we propose a defense method that enhances VLM safety by removing the jailbreak-related shift (JRS-Rem) at inference time. Experiments show that JRS-Rem provides strong defense across multiple scenarios while preserving performance on benign tasks.

Sihao Lin, Pumeng Lyu, Dongrui Liu et al.

Self-attention mechanism is the key of the Transformer but often criticized for its computation demands. Previous token pruning works motivate their methods from the view of computation redundancy but still need to load the full network and require same memory costs. This paper introduces a novel strategy that simplifies vision transformers and reduces computational load through the selective removal of non-essential attention layers, guided by entropy considerations. We identify that regarding the attention layer in bottom blocks, their subsequent MLP layers, i.e. two feed-forward layers, can elicit the same entropy quantity. Meanwhile, the accompanied MLPs are under-exploited since they exhibit smaller feature entropy compared to those MLPs in the top blocks. Therefore, we propose to integrate the uninformative attention layers into their subsequent counterparts by degenerating them into identical mapping, yielding only MLP in certain transformer blocks. Experimental results on ImageNet-1k show that the proposed method can remove 40% attention layer of DeiT-B, improving throughput and memory bound without performance compromise. Code is available at https://github.com/sihaoevery/lambda_vit.

Yu Li, Haoyu Luo, Yuejin Xie et al.

Evaluating the safety of LLM-based agents is increasingly important because risks in realistic deployments often emerge over multi-step interactions rather than isolated prompts or final responses. Existing trajectory-level benchmarks remain limited by insufficient interaction diversity, coarse observability of safety failures, and weak long-horizon realism. We introduce ATBench, a trajectory-level benchmark for structured, diverse, and realistic evaluation of agent safety. ATBench organizes agentic risk along three dimensions: risk source, failure mode, and real-world harm. Based on this taxonomy, we construct trajectories with heterogeneous tool pools and a long-context delayed-trigger protocol that captures realistic risk emergence across multiple stages. The benchmark contains 1,000 trajectories (503 safe and 497 unsafe), averaging 9.01 turns and 3.95k tokens, with 1,954 invoked tools drawn from pools spanning 2,084 available tools. Data quality is supported by rule-based and LLM-based filtering plus full human audit. Experiments on frontier LLMs, open-source models, and specialized guard systems show that ATBench is challenging even for strong evaluators, while enabling taxonomy-stratified analysis, cross-benchmark comparison, and diagnosis of long-horizon failure patterns.

Qihao Lin, Guanxu Chen, Dongrui Liu et al.

As LLMs continue to scale, improving training efficiency increasingly depends on using data more effectively. Data selection addresses this problem by allocating a limited training budget to samples that best promote a target behavior. Existing methods usually represent the target behavior with a set of target examples, but often treat these examples as equally important. This can be inefficient because target examples may differ in their relevance to the current model: examples closer to the model's current behavior provide more actionable guidance than those farther away. We propose PRISM (PReference-aware Influence-function-based Data Selection Method for Efficient Fine-Tuning), which uses the current model's preference to weight target examples and construct a preference-aware target representation. PRISM then scores candidate training samples by their alignment with this representation, concentrating the data budget on samples more likely to move the model toward the target behavior. Theoretical analysis shows that this preference weighting yields a more effective first-order direction for increasing target-behavior preference. Experiments across model families and scales show that PRISM improves both efficient fine-tuning and safety-oriented SFT repair, demonstrating that precise target-behavior characterization is key to budget-efficient data selection.

Chen Qian, Jie Zhang, Wei Yao et al.

Ensuring the trustworthiness of large language models (LLMs) is crucial. Most studies concentrate on fully pre-trained LLMs to better understand and improve LLMs' trustworthiness. In this paper, to reveal the untapped potential of pre-training, we pioneer the exploration of LLMs' trustworthiness during this period, focusing on five key dimensions: reliability, privacy, toxicity, fairness, and robustness. To begin with, we apply linear probing to LLMs. The high probing accuracy suggests that \textit{LLMs in early pre-training can already distinguish concepts in each trustworthiness dimension}. Therefore, to further uncover the hidden possibilities of pre-training, we extract steering vectors from a LLM's pre-training checkpoints to enhance the LLM's trustworthiness. Finally, inspired by~\citet{choi2023understanding} that mutual information estimation is bounded by linear probing accuracy, we also probe LLMs with mutual information to investigate the dynamics of trustworthiness during pre-training. We are the first to observe a similar two-phase phenomenon: fitting and compression~\citep{shwartz2017opening}. This research provides an initial exploration of trustworthiness modeling during LLM pre-training, seeking to unveil new insights and spur further developments in the field. We will make our code publicly accessible at \url{https://github.com/ChnQ/TracingLLM}.

Chen Qian, Dongrui Liu, Haochen Wen et al.

Large reasoning models (LRMs) have demonstrated impressive capabilities in complex problem-solving, yet their internal reasoning mechanisms remain poorly understood. In this paper, we investigate the reasoning trajectories of LRMs from an information-theoretic perspective. By tracking how mutual information (MI) between intermediate representations and the correct answer evolves during LRM reasoning, we observe an interesting MI peaks phenomenon: the MI at specific generative steps exhibits a sudden and significant increase during LRM's reasoning process. We theoretically analyze such phenomenon and show that as MI increases, the probability of model's prediction error decreases. Furthermore, these MI peaks often correspond to tokens expressing reflection or transition, such as ``Hmm'', ``Wait'' and ``Therefore,'' which we term as the thinking tokens. We then demonstrate that these thinking tokens are crucial for LRM's reasoning performance, while other tokens has minimal impacts. Building on these analyses, we propose two simple yet effective methods to improve LRM's reasoning performance, by delicately leveraging these thinking tokens. Overall, our work provides novel insights into the reasoning mechanisms of LRMs and offers practical ways to improve their reasoning capabilities. The code is available at https://github.com/ChnQ/MI-Peaks.

Xiaoya Lu, Dongrui Liu, Yi Yu et al.

Despite the rapid development of safety alignment techniques for LLMs, defending against multi-turn jailbreaks is still a challenging task. In this paper, we conduct a comprehensive comparison, revealing that some existing defense methods can improve the robustness of LLMs against multi-turn jailbreaks but compromise usability, i.e., reducing general capabilities or causing the over-refusal problem. From the perspective of mechanism interpretability of LLMs, we discover that these methods fail to establish a boundary that exactly distinguishes safe and harmful feature representations. Therefore, boundary-safe representations close to harmful representations are inevitably disrupted, leading to a decline in usability. To address this issue, we propose X-Boundary to push harmful representations away from boundary-safe representations and obtain an exact distinction boundary. In this way, harmful representations can be precisely erased without disrupting safe ones. Experimental results show that X-Boundary achieves state-of-the-art defense performance against multi-turn jailbreaks, while reducing the over-refusal rate by about 20% and maintaining nearly complete general capability. Furthermore, we theoretically prove and empirically verify that X-Boundary can accelerate the convergence process during training. Please see our code at: https://github.com/AI45Lab/X-Boundary.

Shaobo Wang, Xiangdong Zhang, Dongrui Liu et al.

Batch Normalization (BN) has become an essential technique in contemporary neural network design, enhancing training stability. Specifically, BN employs centering and scaling operations to standardize features along the batch dimension and uses an affine transformation to recover features. Although standard BN has shown its capability to improve deep neural network training and convergence, it still exhibits inherent limitations in certain cases. Current enhancements to BN typically address only isolated aspects of its mechanism. In this work, we critically examine BN from a feature perspective, identifying feature condensation during BN as a detrimental factor to test performance. To tackle this problem, we propose a two-stage unified framework called Unified Batch Normalization (UBN). In the first stage, we employ a straightforward feature condensation threshold to mitigate condensation effects, thereby preventing improper updates of statistical norms. In the second stage, we unify various normalization variants to boost each component of BN. Our experimental results reveal that UBN significantly enhances performance across different visual backbones and different vision tasks, and notably expedites network training convergence, particularly in early training stages. Notably, our method improved about 3% in accuracy on ImageNet classification and 4% in mean average precision on both Object Detection and Instance Segmentation on COCO dataset, showing the effectiveness of our approach in real-world scenarios.

Zhonghao Yang, Yu Li, Yanxu Zhu et al.

As agent systems move into increasingly diverse execution settings, trajectory-level safety evaluation and diagnosis require benchmarks that evolve with them. ATBench is a diverse and realistic agent trajectory benchmark for safety evaluation and diagnosis. This report presents ATBench-Claw and ATBench-CodeX, two domain-customized extensions that carry ATBench into the OpenClaw and OpenAI Codex / Codex-runtime settings. The key adaptation mechanism is to analyze each new setting, customize the three-dimensional Safety Taxonomy over risk source, failure mode, and real-world harm, and then use that customized taxonomy to define the benchmark specification consumed by the shared ATBench construction pipeline. This extensibility matters because agent frameworks remain relatively stable at the architectural level even as their concrete execution settings, tool ecosystems, and product capabilities evolve quickly. Concretely, ATBench-Claw targets OpenClaw-sensitive execution chains over tools, skills, sessions, and external actions, while ATBench-CodeX targets trajectories in the OpenAI Codex / Codex-runtime setting over repositories, shells, patches, dependencies, approvals, and runtime policy boundaries. Our emphasis therefore falls on taxonomy customization, domain-specific risk coverage, and benchmark design under a shared ATBench generation framework.

Zichen Wen, Jiashu Qu, Dongrui Liu et al.

Diffusion-based large language models (dLLMs) have recently emerged as a powerful alternative to autoregressive LLMs, offering faster inference and greater interactivity via parallel decoding and bidirectional modeling. However, despite strong performance in code generation and text infilling, we identify a fundamental safety concern: existing alignment mechanisms fail to safeguard dLLMs against context-aware, masked-input adversarial prompts, exposing novel vulnerabilities. To this end, we present DIJA, the first systematic study and jailbreak attack framework that exploits unique safety weaknesses of dLLMs. Specifically, our proposed DIJA constructs adversarial interleaved mask-text prompts that exploit the text generation mechanisms of dLLMs, i.e., bidirectional modeling and parallel decoding. Bidirectional modeling drives the model to produce contextually consistent outputs for masked spans, even when harmful, while parallel decoding limits model dynamic filtering and rejection sampling of unsafe content. This causes standard alignment mechanisms to fail, enabling harmful completions in alignment-tuned dLLMs, even when harmful behaviors or unsafe instructions are directly exposed in the prompt. Through comprehensive experiments, we demonstrate that DIJA significantly outperforms existing jailbreak methods, exposing a previously overlooked threat surface in dLLM architectures. Notably, our method achieves up to 100% keyword-based ASR on Dream-Instruct, surpassing the strongest prior baseline, ReNeLLM, by up to 78.5% in evaluator-based ASR on JailbreakBench and by 37.7 points in StrongREJECT score, while requiring no rewriting or hiding of harmful content in the jailbreak prompt. Our findings underscore the urgent need for rethinking safety alignment in this emerging class of language models. Code is available at https://github.com/ZichenWen1/DIJA.

Shiyang Feng, Runmin Ma, Xiangchao Yan et al.

We introduce InternAgent-1.5, a unified system designed for end-to-end scientific discovery across computational and empirical domains. The system is built on a structured architecture composed of three coordinated subsystems for generation, verification, and evolution. These subsystems are supported by foundational capabilities for deep research, solution optimization, and long horizon memory. The architecture allows InternAgent-1.5 to operate continuously across extended discovery cycles while maintaining coherent and improving behavior. It also enables the system to coordinate computational modeling and laboratory experimentation within a single unified system. We evaluate InternAgent-1.5 on scientific reasoning benchmarks such as GAIA, HLE, GPQA, and FrontierScience, and the system achieves leading performance that demonstrates strong foundational capabilities. Beyond these benchmarks, we further assess two categories of discovery tasks. In algorithm discovery tasks, InternAgent-1.5 autonomously designs competitive methods for core machine learning problems. In empirical discovery tasks, it executes complete computational or wet lab experiments and produces scientific findings in earth, life, biological, and physical domains. Overall, these results show that InternAgent-1.5 provides a general and scalable framework for autonomous scientific discovery.

Dadi Guo, Qingyu Liu, Dongrui Liu et al.

Large Language Model (LLM)-based agents are increasingly used as autonomous subordinates that carry out tasks for users. This raises the question of whether they may also engage in deception, similar to how individuals in human organizations lie to superiors to create a good image or avoid punishment. We observe and define agentic upward deception, a phenomenon in which an agent facing environmental constraints conceals its failure and performs actions that were not requested without reporting. To assess its prevalence, we construct a benchmark of 200 tasks covering five task types and eight realistic scenarios in a constrained environment, such as broken tools or mismatched information sources. Evaluations of 11 popular LLMs reveal that these agents typically exhibit action-based deceptive behaviors, such as guessing results, performing unsupported simulations, substituting unavailable information sources, and fabricating local files. We further test prompt-based mitigation and find only limited reductions, suggesting that it is difficult to eliminate and highlighting the need for stronger mitigation strategies to ensure the safety of LLM-based agents.

Peiliang Cai, Evelyn Zhang, Jiacheng Liu et al.

Recent advances in autoregressive video diffusion have enabled sequential and streaming video generation. However, long-horizon generation requires increasingly large KV caches, making efficient compression without sacrificing quality challenging. Existing methods mostly select historical frames based on attention scores, but their context decisions remain coarse. When multiple frames are generated in the same chunk, these methods often apply a shared history selection to the whole chunk, score historical frames solely by attention, and assign head-wise budgets either uniformly or by attention-pattern heuristics rather than explicit head-importance estimation. We show that frames within the same generated chunk can depend on distinct historical frames, that the same historical frame can receive different attention scores as its relative temporal distance to the current frames changes, and that masking different heads induces unequal generation degradation. Motivated by these findings, we propose \textbf{Focused Forcing}, a training-free KV selection method that focuses cached history along both generated-frame and head dimensions. For each generated frame, Focused Forcing preserves the most relevant and distinctive historical frames by combining attention scores with diversity scores of historical frames, while assigning larger budgets to heads with higher estimated importance. Across multiple autoregressive generation paradigms, Focused Forcing achieves up to $\textbf{1.48}\times$ end-to-end acceleration without training, while \textbf{improving visual quality and text alignment}. \textit{Our code will be released on GitHub.}

Junyao Yang, Chen Qian, Kun Wang et al.

The advancement of Large Reasoning Models (LRMs) has catalyzed a paradigm shift from reactive ``fast thinking'' text generation to systematic, step-by-step ``slow thinking'' reasoning, unlocking state-of-the-art performance in complex mathematical and logical tasks. However, the field faces \textit{the fundamental gap between token-level behavioral analysis and internal reasoning mechanisms, and the instability of reinforcement learning (RL) for reasoning optimization relying on costly external verifiers}. We identify and formally define \textbf{Entropy-Gradient Inversion}, a robust negative correlation between token entropy and logit gradients that acts as a definitive geometric fingerprint for LRM reasoning capability. Building on this, we propose \textbf{Correlation-Regularized Group Policy Optimization (CorR-PO)}, which embeds this inversion signature into RL reward regularization. Extensive experiments on various reasoning benchmarks across multiple model scales show CorR-PO consistently outperforms state-of-the-art baselines, confirming that stronger inversion directly correlates with superior reasoning performance.

Guanxu Chen, Dongrui Liu, Jing Shao

Large Language Models (LLMs) often exhibit a gap between their internal knowledge and their explicit linguistic outputs. In this report, we empirically investigate whether Looped Transformers (LTs)--architectures that increase computational depth by iterating shared layers--can bridge this gap by utilizing their iterative nature as a form of introspection. Our experiments reveal that while increasing loop iterations narrows the gap, it is partly driven by a degradation of their internal knowledge carried by representations. Moreover, another empirical analysis suggests that current LTs' ability to perceive representations does not improve across loops; it is only present in the final loop. These results suggest that while LTs offer a promising direction for scaling computational depth, they have yet to achieve the introspection required to truly link representation space and natural language.

Qihan Ren, Peng Wang, Ruikun Cai et al.

A prevailing narrative in LLM post-training holds that supervised finetuning (SFT) memorizes while reinforcement learning (RL) generalizes. We revisit this claim for reasoning SFT with long chain-of-thought (CoT) supervision and find that cross-domain generalization is not absent but conditional, jointly shaped by optimization dynamics, training data, and base-model capability. Some reported failures are under-optimization artifacts: cross-domain performance first degrades before recovering and improving with extended training (a dip-and-recovery pattern), so shorttraining checkpoints can underestimate generalization. Data quality and structure both matter: low-quality solutions broadly hurt generalization,while verified long-CoT traces yield consistent cross-domain gains. Model capability is essential: stronger models internalize transferable procedural patterns (e.g., backtracking) even from a toy arithmetic game, while weaker ones imitate surface verbosity. This generalization is asymmetric, however: reasoning improves while safety degrades, reframing the question from whether reasoning SFT generalizes to under what conditions and at what cost.

Zhiqiang Wang, Dongrui Liu, Yan Li et al.

Existing adversarial attacks on vision-language models (VLMs) can steer model outputs toward attacker-specified target responses, but their effectiveness often degrades when the same perturbed input is paired with different textual queries. This paper studies cross-query response manipulation, where a single adversarial example is expected to remain effective across diverse user queries. We first analyze the limitations of existing attacks and find that successful transfer is closely associated with preserving an image-dominant attention pattern during response generation. Motivated by the observation, we propose \textbf{Attention Hijacking}, a novel adversarial attack that explicitly steers internal attention distributions toward a persistent image-dominant pattern. By amplifying the influence of visual tokens on target response tokens while suppressing the competing influence of textual tokens, our method reduces the dependence of the manipulated output on the specific wording of the query. Extensive experiments on widely used VLMs show that Attention Hijacking substantially improves cross-query transferability across diverse target responses and unseen queries. The method also extends effectively to multiple attack scenarios, offering new insights into the role of attention stability in transferable response manipulation for VLMs.

Xiaoya Lu, Zeren Chen, Xuhao Hu et al.

Flawed planning from VLM-driven embodied agents poses significant safety hazards, hindering their deployment in real-world household tasks. However, existing static, non-interactive evaluation paradigms fail to adequately assess risks within these interactive environments, since they cannot simulate dynamic risks that emerge from an agent's actions and rely on unreliable post-hoc evaluations that ignore unsafe intermediate steps. To bridge this critical gap, we propose evaluating an agent's interactive safety: its ability to perceive emergent risks and execute mitigation steps in the correct procedural order. We thus present IS-Bench, the first multi-modal benchmark designed for interactive safety, featuring 161 challenging scenarios with 388 unique safety risks instantiated in a high-fidelity simulator. Crucially, it facilitates a novel process-oriented evaluation that verifies whether risk mitigation actions are performed before/after specific risk-prone steps. Extensive experiments on leading VLMs, including the GPT-4o and Gemini-2.5 series, reveal that current agents lack interactive safety awareness, and that while safety-aware Chain-of-Thought can improve performance, it often compromises task completion. By highlighting these critical limitations, IS-Bench provides a foundation for developing safer and more reliable embodied AI systems. Code and data are released under [this https URL](https://github.com/AI45Lab/IS-Bench).

Xiangyang Zhu, Yuan Tian, Qi Jia et al.

The success of large language models (LLMs) in scientific domains has heightened safety concerns, prompting numerous benchmarks to evaluate their scientific safety. Existing benchmarks often suffer from limited risk coverage and a reliance on subjective evaluation. To address these problems, we introduce SafeSci, a comprehensive framework for safety evaluation and enhancement in scientific contexts. SafeSci comprises SafeSciBench, a multi-disciplinary benchmark with 0.25M samples, and SafeSciTrain, a large-scale dataset containing 1.5M samples for safety enhancement. SafeSciBench distinguishes between safety knowledge and risk to cover extensive scopes and employs objective metrics such as deterministically answerable questions to mitigate evaluation bias. We evaluate 24 advanced LLMs, revealing critical vulnerabilities in current models. We also observe that LLMs exhibit varying degrees of excessive refusal behaviors on safety-related issues. For safety enhancement, we demonstrate that fine-tuning on SafeSciTrain significantly enhances the safety alignment of models. Finally, we argue that knowledge is a double-edged sword, and determining the safety of a scientific question should depend on specific context, rather than universally categorizing it as safe or unsafe. Our work provides both a diagnostic tool and a practical resource for building safer scientific AI systems.

Junyao Yang, Chen Qian, Dongrui Liu et al.

Large Reasoning Models (LRMs) with long chain-of-thought reasoning have recently achieved remarkable success. Yet, equipping domain-specialized models with such reasoning capabilities, referred to as "Reasoning + X", remains a significant challenge. While model merging offers a promising training-free solution, existing methods often suffer from a destructive performance collapse: existing methods tend to both weaken reasoning depth and compromise domain-specific utility. Interestingly, we identify a counter-intuitive phenomenon underlying this failure: reasoning ability predominantly resides in parameter regions with low gradient sensitivity, contrary to the common assumption that domain capabilities correspond to high-magnitude parameters. Motivated by this insight, we propose ReasonAny, a novel merging framework that resolves the reasoning-domain performance collapse through Contrastive Gradient Identification. Experiments across safety, biomedicine, and finance domains show that ReasonAny effectively synthesizes "Reasoning + X" capabilities, significantly outperforming state-of-the-art baselines while retaining robust reasoning performance.

Lige Huang, Zicheng Liu, Jie Zhang et al.

The dual offensive and defensive utility of Large Language Models (LLMs) highlights a critical gap in AI security: the lack of unified frameworks for dynamic, iterative adversarial adaptation hardening. To bridge this gap, we propose the Red Team vs. Blue Team (RvB) framework, formulated as a training-free, sequential, imperfect-information game. In this process, the Red Team exposes vulnerabilities, driving the Blue Team to learning effective solutions without parameter updates. We validate our framework across two challenging domains: dynamic code hardening against CVEs and guardrail optimization against jailbreaks. Our empirical results show that this interaction compels the Blue Team to learn fundamental defensive principles, leading to robust remediations that are not merely overfitted to specific exploits. RvB achieves Defense Success Rates of 90\% and 45\% across the respective tasks while maintaining near 0\% False Positive Rates, significantly surpassing baselines. This work establishes the iterative adversarial interaction framework as a practical paradigm that automates the continuous hardening of AI systems.

Ling Tang, Jilin Mei, Dongrui Liu et al.

Large language model-powered multi-agent systems have emerged as powerful tools for simulating complex human-like systems. The interactions within these systems often lead to extreme events whose origins remain obscured by the black box of emergence. Interpreting these events is critical for system safety. This paper proposes the first framework for explaining emergent extreme events in multi-agent systems, aiming to answer three fundamental questions: When does the event originate? Who drives it? And what behaviors contribute to it? Specifically, we adapt the Shapley value to faithfully attribute the occurrence of extreme events to each action taken by agents at different time steps, i.e., assigning an attribution score to the action to measure its influence on the event. We then aggregate the attribution scores along the dimensions of time, agent, and behavior to quantify the risk contribution of each dimension. Finally, we design a set of metrics based on these contribution scores to characterize the features of extreme events. Experiments across diverse multi-agent system scenarios (economic, financial, and social) demonstrate the effectiveness of our framework and provide general insights into the emergence of extreme phenomena.

Qianli Ma, Dongrui Liu, Qian Chen et al.

Fine-tuning pre-trained Large Language Models (LLMs) for specialized tasks incurs substantial computational and data costs. While model merging offers a training-free solution to integrate multiple task-specific models, existing methods suffer from safety-utility conflicts where enhanced general capabilities degrade safety safeguards. We identify two root causes: $\textbf{neuron misidentification}$ due to simplistic parameter magnitude-based selection, and $\textbf{cross-task neuron interference}$ during merging. To address these challenges, we propose $\textbf{LED-Merging}$, a three-stage framework that $\textbf{L}$ocates task-specific neurons via gradient-based attribution, dynamically $\textbf{E}$lects critical neurons through multi-model importance fusion, and $\textbf{D}$isjoints conflicting updates through parameter isolation. Extensive experiments on Llama-3-8B, Mistral-7B, and Llama2-13B demonstrate that LED-Merging effectively reduces harmful response rates, showing a 31.4\% decrease on Llama-3-8B-Instruct on HarmBench, while simultaneously preserving 95\% of utility performance, such as achieving 52.39\% accuracy on GSM8K. LED-Merging resolves safety-utility conflicts and provides a lightweight, training-free paradigm for constructing reliable multi-task LLMs. Code is available at $\href{https://github.com/MqLeet/LED-Merging}{GitHub}$.

Chen Qian, Peng Wang, Dongrui Liu et al.

Large Language Model (LLM)-based agents are widely used in real-world applications such as customer service, web navigation, and software engineering. As these systems become more autonomous and are deployed at scale, understanding why an agent takes a particular action becomes increasingly important for accountability and governance. However, existing research predominantly focuses on \textit{failure attribution} to localize explicit errors in unsuccessful trajectories, which is insufficient for explaining the reasoning behind agent behaviors. To bridge this gap, we propose a novel framework for \textbf{general agentic attribution}, designed to identify the internal factors driving agent actions regardless of the task outcome. Our framework operates hierarchically to manage the complexity of agent interactions. Specifically, at the \textit{component level}, we employ temporal likelihood dynamics to identify critical interaction steps; then at the \textit{sentence level}, we refine this localization using perturbation-based analysis to isolate the specific textual evidence. We validate our framework across a diverse suite of agentic scenarios, including standard tool use and subtle reliability risks like memory-induced bias. Experimental results demonstrate that the proposed framework reliably pinpoints pivotal historical events and sentences behind the agent behavior, offering a critical step toward safer and more accountable agentic systems.

Yue Xu, Qian Chen, Zizhan Ma et al.

Large language models have enabled agents that reason, plan, and interact with tools and environments to accomplish complex tasks. As these agents operate over extended interaction horizons, their effectiveness increasingly depends on adapting behavior to individual users and maintaining continuity across time, giving rise to personalized LLM-powered agents. In such long-term, user-dependent settings, personalization permeates the entire decision pipeline rather than remaining confined to surface-level generation. This survey provides a capability-oriented review of personalized LLM-powered agents. We organize the literature around four interdependent components: profile modeling, memory, planning, and action execution. Using this taxonomy, we synthesize representative methods and analyze how user signals are represented, propagated, and utilized, highlighting cross-component interactions and recurring design trade-offs. We further examine evaluation metrics and benchmarks tailored to personalized agents, summarize application scenarios spanning general assistance to specialized domains, and outline future directions for research and deployment. By offering a structured framework for understanding and designing personalized LLM-powered agents, this survey charts a roadmap toward more user-aligned, adaptive, robust, and deployable agentic systems, accelerating progress from prototype personalization to scalable real-world assistants.

Yijin Zhou, Xiaoya Lu, Dongrui Liu et al.

The rapid advancement of Large Language Model (LLM)-based Multi-Agent Systems (MAS) has introduced significant security vulnerabilities, where malicious influence can propagate virally through inter-agent communication. Conventional safeguards often rely on a binary paradigm that strictly distinguishes between benign and attack agents, failing to account for infected agents i.e., benign entities converted by attack agents. In this paper, we propose Infection-Aware Guard, INFA-Guard, a novel defense framework that explicitly identifies and addresses infected agents as a distinct threat category. By leveraging infection-aware detection and topological constraints, INFA-Guard accurately localizes attack sources and infected ranges. During remediation, INFA-Guard replaces attackers and rehabilitates infected ones, avoiding malicious propagation while preserving topological integrity. Extensive experiments demonstrate that INFA-Guard achieves state-of-the-art performance, reducing the Attack Success Rate (ASR) by an average of 33%, while exhibiting cross-model robustness, superior topological generalization, and high cost-effectiveness.

Jianuo Huang, Yaojie Zhang, Yicun Yang et al. · cmu

Diffusion large language models (dLLMs) present a promising alternative to dominant autoregressive models (ARMs) by the ability of parallel decoding at the expense of substantial computation and memory costs. Specifically, the cache mechanism for bidirectional attention in dLLMs demands large memory footprint, restricting their ability to handle long contexts under resource-limited settings. Existing cache eviction strategies are designed for ARMs and ignore the unique characteristics of dLLMs, thus leading to unsatisfactory performance. To address these challenges, we introduce MaskKV, a training-free cache eviction framework tailored to dLLMs, focusing on the effect of mask tokens in dLLMs. MaskKV is built on two key innovations: (1) a mask-query guided scoring mechanism that leverages attention weights to identify and evict less critical prompt tokens for each head; (2) an adaptive cache budgeting strategy that improves efficiency by reducing allocation in intermediate layers and concentrating resources on prompt-preferring heads. On LLaDA with MaskKV, compressing the KV cache to only 256 pairs (less than 5% of tokens) retains 94% of the full-cache performance on LongBench and achieves up to 31x acceleration at 32k prompt length. The code is publicly available at: https://github.com/jianuo-huang/MaskKV

Shuai Shao, Qihan Ren, Chen Qian et al. · princeton

Advances in Large Language Models (LLMs) have enabled a new class of self-evolving agents that autonomously improve through interaction with the environment, demonstrating strong capabilities. However, self-evolution also introduces novel risks overlooked by current safety research. In this work, we study the case where an agent's self-evolution deviates in unintended ways, leading to undesirable or even harmful outcomes. We refer to this as Misevolution. To provide a systematic investigation, we evaluate misevolution along four key evolutionary pathways: model, memory, tool, and workflow. Our empirical findings reveal that misevolution is a widespread risk, affecting agents built even on top-tier LLMs (e.g., Gemini-2.5-Pro). Different emergent risks are observed in the self-evolutionary process, such as the degradation of safety alignment after memory accumulation, or the unintended introduction of vulnerabilities in tool creation and reuse. To our knowledge, this is the first study to systematically conceptualize misevolution and provide empirical evidence of its occurrence, highlighting an urgent need for new safety paradigms for self-evolving agents. Finally, we discuss potential mitigation strategies to inspire further research on building safer and more trustworthy self-evolving agents. Our code and data are available at https://github.com/ShaoShuai0605/Misevolution . Warning: this paper includes examples that may be offensive or harmful in nature.