The evolution of social bots into highly realistic human imitators renders traditional detection methods, which primarily rely on static content or metadata, increasingly ineffective. This necessitates a paradigm shift towards analyzing deep behavioral attributes, which reveal the intrinsic regularities that remain difficult to disguise even with advanced generative capabilities. To achieve this, we propose a novel framework that models behavior from the perspective of user intention to explicitly mine underlying camouflage patterns. Specifically, we introduce a novel intent-aware encoding paradigm that synergizes explicit actions with latent user motives to capture fine-grained behavioral logic. To tackle the heterogeneity of bot strategies, we devise a prototype-based graph contrastive learning mechanism that disentangles camouflaged bots from genuine users in the feature space. Furthermore, an adaptive collaboration inference module is implemented to dynamically calibrate the model’s reliance on behavioral versus non-behavioral signals in a dual-channel framework. Extensive experiments on Cresci-2015 (5,301 users), TwiBot-20 (229,580 users), and TwiBot-22 (1 million users) show consistent gains over strong baselines. Our method achieves F1-scores of 98.97%, 89.33%, and 59.07% respectively, achieving state-of-the-art performance across datasets reflecting different stages of bot evolution.

Diffusion Transformers (DiTs) excel at generation, but their global self-attention makes controllable, reference-image-based editing a distinct challenge. Unlike U-Nets, naively injecting local appearance into a DiT can disrupt its holistic scene structure. We address this by proposing the first training-free framework specifically designed to tame DiTs for high-fidelity appearance transfer. Our core is a synergistic system that disentangles structure and appearance. We leverage high-fidelity inversion to establish a rich content prior for the source image, capturing its lighting and micro-textures. A novel attention-sharing mechanism then dynamically fuses purified appearance features from a reference, guided by geometric priors. Our unified approach operates at 1024px and outperforms specialized methods on tasks ranging from semantic attribute transfer to fine-grained material application. Extensive experiments confirm our state-of-the-art performance in both structural preservation and appearance fidelity.

Despite the success of diffusion models on text-to-image identity customization, their misuse can exacerbate the generation of misleading and harmful content, endangering individuals and society. To address this problem, existing methods add protective perturbations into user images, which can degrade the quality and fidelity of customized images. We identify two limitations of these protection methods: i) lack of universal protection efficacy against both tuning-based and tuning-free customization models, and ii) poor balance between protection efficacy and cost. To address these issues, we propose Targeted Ensemble Defense (TED), a method that can defend against both tuning-based and tuning-free models. Specifically, we employ an ensemble of both tuning-based and tuning-free models to craft universal perturbations. Furthermore, by leveraging the targeted perturbations to disrupt the image encoders, we reduce protection cost without sacrificing protection efficacy. TED outperforms the SOTA method by 41.49% of FDFR scores and 23.94% of ISM scores across four customization models and two facial benchmarks on average. Notably, TED can successfully protect ID information in user images from unseen base models and customization methods, demonstrating its excellent transferability.

Multimodal Misinformation Detection (MMD) in realistic, mixed-sourced scenarios must incorporate robust reasoning capabilities to handle the social complexity and diverse types of forgeries. While MLLM-based agents are increasingly used for MMD task due to their powerful reasoning abilities, they suffer from a critical trade-off: on one hand, single-agent methods provide only the limited, single-view analysis; on the other hand, multi-agent methods introduce high computational costs and significant optimization difficulties. To address this gap, we propose a novel Multi-Social-Agent Self-Distillation framework that internalizes collective social reasoning capabilities into a unified model. Our framework consists of two core stages: (1) we simulate multi-perspective judgments from a diverse society of MLLM agents and synthesize their collective feedback into high-quality Social Chain-of-Thought (SCoT) data; (2) Building on this, we propose the Social Correction Value-Driven Preference Optimization (SCPO), a new alignment algorithm that leverages the degree of social misjudgment as a verifiable signal to dynamically focus training on the most challenging samples. Extensive experiments on the challenging MFC-Bench and MMFakeBench benchmarks demonstrate the effectiveness of our framework. Our 7B Qwen2-VL-based model significantly outperforms various MLLM baselines, multi-agent methods, and even competes or surpasses proprietary models like GPT-4o and Claude, facilitating advanced multimodal misinformation reasoning and detection via thinking as society.

Online advertising is central to e-commerce, and the wording of advertising texts influences consumer behavior and sales. Most existing methods optimize for click through rate (CTR) or conversion rate (CVR), which focus on final outcomes and do not reflect how consumers make decisions at different stages. We study advertising text generation conditioned on product images and textual descriptions, and propose a framework called Social Simulator-Guided Reinforcement Learning (SSG-RL). A social simulator built on large language models groups consumers using the VALS-2 (Values, Attitudes, and Lifestyles) model and scores advertising texts along the McKinsey Consumer Decision Journey (CDJ). These stage aware scores are used to train a reward model and to guide reinforcement learning for generating improved advertising texts. Experiments on an e-commerce dataset show that SSG-RL outperforms strong baselines on several key metrics. Our method improves the fluency, diversity, and image–text alignment of advertising texts by 10.6%, 3.8%, and 5.4%, respectively. It also encourages paraphrasing rather than direct copying, which may lead to a slight drop in lexical-overlap metrics such as BERTScore. Furthermore, a human study with 190 consumers and 3 experts shows that our method is preferred in pairwise comparisons, with 18.98% and 6.67% higher preference rates.

Sequential recommendation is a classic task in the field of recommendation, which aims to predict the next user-preferred item based on their historical interactions. However, in practical scenarios, users' needs are dynamically evolving in a short period and exhibit a chain-like structure. Consequently, recommending only the next single item does not fully meet user demands and limits the potential for increasing business traffic on platforms. To overcome this limitation, we propose a new recommendation paradigm, Next Chain Prediction, which requires the model to predict a chain of items. Due to the advantages of generative recommendation models on user preference representation and scalability, we design a generative recommendation model for next chain prediction. The generative model extracts long-term interests and short-term demands within a unified framework. By designing a Sequence-Chain Attention mechanism, the model performs self-attention learning across dual dimensions. Additionally, we design a generative loss function to balance the hit rate and diversity of the recommended items in the chain. We conduct experiments across three datasets and the experimental results show that our method achieves at least 1.22% improvement in HR@10 across three datasets for recommending multi-item chains. Furthermore, our method improves the diversity of recommended items and also offers the flexibility to adjust the size of predicted chains, maintaining state-of-the-art performance even when limited to predicting a single item.

Recent studies on LLM agent scaling have highlighted the potential of Multi-Agent Debate (MAD) to enhance reasoning abilities. However, the critical aspect of role allocation strategies remains underexplored. In this study, we demonstrate that allocating roles with differing viewpoints to specific positions significantly impacts MAD's performance in reasoning tasks. Specifically, we find a novel role allocation strategy, "Truth Last", which can improve MAD performance by up to 22% in reasoning tasks. To address the issue of unknown truth in practical applications, we propose the Multi-Agent Debate Consistency (MADC) strategy, which systematically simulates and optimizes its core mechanisms. MADC incorporates path consistency to assess agreement among independent roles, simulating the role with the highest consistency score as the truth. We validated MADC across a range of LLMs (9 models), including the DeepSeek-R1 Distilled Models, on challenging reasoning tasks. MADC consistently demonstrated advanced performance, effectively overcoming MAD's performance bottlenecks and providing a crucial pathway for further improvements in LLM agent scaling.

Large language models (LLMs) have demonstrated exceptional proficiency in language understanding. However, when LLMs align their outputs with deceptive and/or misleading prompts, the generated responses could deviate from the de facto information. Such observations are known as fawning hallucinations, where the model prioritizes alignment with the input's implied perspective over accuracy and truthfulness. In this work, we analyze fawning hallucinations in various natural language processing tasks and tailor the so-termed contrastive decoding method for fawning-hallucination mitigation. Specifically, we design two paradigms to generate corresponding deceptive and/or misleading inputs for the consistent fawning hallucinations induction. Then, we propose the collaborative contrastive decoding (CCD) to handle the fawning hallucinations across different tasks in LLMs. By contrasting the deviation in output distribution between induced and transformed neutral inputs, the proposed CCD can reduce reliance on deceptive and/or misleading information without requiring additional training. Extensive experiments demonstrate that the proposed CCD can effectively mitigate fawning hallucinations and improve the factuality of the generated responses over various tasks.

With breakthroughs in deep generative technologies such as diffusion models,visual generation models have achieved significant leaps in generation quality and semantic consistency,finding extensive applications in fields like artistic creation and industrial design.However,the powerful generative capability has also triggered severe content safety risks.Malicious users can induce models to generate pornographic,violent,or copyright-infringing images,posing an urgent need for the safety governance of generative AI.This paper provides a systematic review that focuses on two core adversarial tasks of T2I models:(1)Jailbreak attacks,which aim to induce models to breach safety guardrails;(2)Concept erasure,which aims to eliminate internal risk knowledge from the models.First,we establish a taxonomy of jailbreak attacks.By analyzing them across four dimensions:Technical category,perturbation strategy,query type,and adversary knowledge,we reveal the evolutionary trend of attack methods shifting from feature-space perturbations to semantic-space reasoning.Second,regarding risk governance,this paper delves into concept erasure technologies,comparatively analyzing three mainstream technical routes:Model fine-tuning,model editing,and inference guidance.We elucidate the trade-offs among erasure effectiveness,computational efficiency,and the preservation of general generation capabilities.Finally,we summarize the commonly used benchmark datasets in this field and identify the current challenges and future directions regarding adversarial robustness and multi-concept joint governance,aiming to provide theoretical references and technical guidance for building safe and controllable T2I systems.

Large Reasoning Models (LRMs) have demonstrated strong capabilities in generating step-by-step reasoning chains alongside final answers, enabling their deployment in high-stakes domains such as healthcare and education. While prior jailbreak attack studies have focused on the safety of final answers, little attention has been given to the safety of the reasoning process. In this work, we identify a novel problem that injects harmful content into the reasoning steps while preserving unchanged answers. This type of attack presents two key challenges: 1) manipulating the input instructions may inadvertently alter the LRM's final answer, and 2) the diversity of input questions makes it difficult to consistently bypass the LRM's safety alignment mechanisms and embed harmful content into its reasoning process. To address these challenges, we propose the Psychology-based Reasoning-targeted Jailbreak Attack (PRJA) Framework, which integrates a Semantic-based Trigger Selection module and a Psychology-based Instruction Generation module. Specifically, the proposed PRJA automatically selects manipulative reasoning triggers via semantic analysis and leverages psychological theories of obedience to authority and moral disengagement to generate adaptive instructions for enhancing the LRM's compliance with harmful content generation. Extensive experiments on five question-answering datasets demonstrate that PRJA achieves an average attack success rate of 83.6% against several commercial LRMs, including DeepSeek R1, Qwen2.5-Max, and OpenAI o4-mini.

Physics-aware symbolic simulation of 3D scenes is critical for robotics, embodied AI, and scientific computing, requiring models to understand natural language descriptions of physical phenomena and translate them into executable simulation environments. While large language models (LLMs) excel at general code generation, they struggle with the semantic gap between physical descriptions and simulation implementation. We introduce PhysCodeBench, the first comprehensive benchmark for evaluating physics-aware symbolic simulation, comprising 700 manually-crafted diverse samples across mechanics, fluid dynamics, and soft-body physics with expert annotations. Our evaluation framework measures both code executability and physical accuracy through automated and visual assessment. Building on this, we propose a Self-Corrective Multi-Agent Refinement Framework (SMRF) with three specialized agents (simulation generator, error corrector, and simulation refiner) that collaborate iteratively with domain-specific validation to produce physically accurate simulations. SMRF achieves 67.7 points overall performance compared to 36.3 points for the best baseline among evaluated SOTA models, representing a 31.4-point improvement. Our analysis demonstrates that error correction is critical for accurate physics-aware symbolic simulation and that specialized multi-agent approaches significantly outperform single-agent methods across the tested physical domains.

Sea surface temperature (SST) forecasting is fundamental to understanding ocean-atmosphere interactions and maintaining marine ecosystem stability. SST exhibits complex spatiotemporal dynamics characterized by two inherent challenges: global-scale models often fail to resolve local variability, while the gradual nature of global changes obscures short-term temporal fluctuations. Existing approaches exhibit limited capability in simultaneously resolving fine-scale local features and capturing dynamic spatiotemporal evolution patterns. To address these dual challenges in SST forecasting, we propose a novel global daily prediction framework with a two-day temporal resolution, combining weather foundation model adapter tuning with physics-informed optimization. The architecture employs a hybrid 3D-ConvNext design that hierarchically integrates volumetric convolutional representations for multiscale spatiotemporal learning. For temporal modeling, we implement a physics-constrained optimization strategy featuring differential gradient regularization for slow variations and spectral-domain enhancement for transient fluctuations. Comprehensive experiments demonstrate that our framework achieves state-of-the-art performance in global SST forecasting while showing particular improvements in resolving challenging local dynamics across diverse marine regimes.

Using risky text prompts, such as pornography and violent prompts, to test the safety of text-to-image (T2I) models is a critical task. However, existing risky prompt datasets are limited in three key areas: 1) limited risky categories, 2) coarse-grained annotation, and 3) low effectiveness. To address these limitations, we introduce T2I-RiskyPrompt, a comprehensive benchmark designed for evaluating safety-related tasks in T2I models. Specifically, we first develop a hierarchical risk taxonomy, which consists of 6 primary categories and 14 fine-grained subcategories. Building upon this taxonomy, we construct a pipeline to collect and annotate risky prompts. Finally, we obtain 6,432 effective risky prompts, where each prompt is annotated with both hierarchical category labels and detailed risk reasons. Moreover, to facilitate the evaluation, we propose a reason-driven risky image detection method that explicitly aligns the MLLM with safety annotations. Based on T2I-RiskyPrompt, we conduct a comprehensive evaluation of eight T2I models, nine defense methods, five safety filters, and five attack strategies, offering nine key insights into the strengths and limitations of T2I model safety. Finally, we discuss potential applications of T2I-RiskyPrompt across various research fields.

Large language model (LLM)-based multi-agent systems have experienced rapid growth in recent years. Despite their promise, such systems remain fragile, frequently exhibiting reasoning and coordination errors that can lead to system-level failures. Failure attribution in such systems relies on tracing natural language interactions among agents to identify the decisive error, which refers to the earliest action whose correction can reverse system failure. There are two key challenges: 1) Shallow attribution: Existing methods often capture only minor deviations, such as incomplete retrievals or formatting errors, which verification mechanisms can correct, while missing the decisive cause of system failure. 2) Contextual degradation: As the length of the system traces increases, the model's reasoning ability rapidly deteriorates. To address these challenges, we propose DCFA, a training-free framework for failure attribution. DCFA integrates a global module that constructs structured causal-inspired dependency graphs from system traces to identify the initial decisive error, and a local module that applies local counterfactual-inspired reasoning to refine causal-inspired attribution. Experiments on the Who When benchmark across six LLMs show that DCFA improves step-level accuracy by up to 8.27

Region-instructed layout control in text-to-image generation is highly practical, yet existing methods suffer from limitations: (i) training-based approaches inherit data bias and often degrade image quality, and (ii) current techniques struggle with occlusion order, limiting real-world usability. To address these issues, we propose LayerBind. By modeling regional generation as distinct layers and binding them during the generation, our method enables precise regional and occlusion controllability. Our motivation stems from the observation that spatial layout and occlusion are established at a very early denoising stage, suggesting that rearranging the early latent structure is sufficient to modify the final output. Building on this, we structure the scheme into two phases: instance initialization and subsequent semantic nursing. (1) First, leveraging the contextual sharing mechanism in multimodal joint attention, Layer-wise Instance Initialization creates per-instance branches that attend to their own regions while anchoring to the shared background. At a designated early step, these branches are fused according to the layer order to form a unified latent with a pre-established layout. (2) Then, Layer-wise Semantic Nursing reinforces regional details and maintains the occlusion order via a layer-wise attention enhancement. Specifically, a sequential layered attention path operates alongside the standard global path, with updates composited under a layer-transparency scheduler. LayerBind is training-free and plug-and-play, serving as a regional and occlusion controller across Diffusion Transformers. Beyond generation, it natively supports editable workflows, allowing for flexible modifications like changing instances or rearranging visible orders. Both qualitative and quantitative results demonstrate LayerBind's effectiveness, highlighting its strong potential for creative applications.

Autonomous Graphical User Interface (GUI) agents often struggle with multi-step tasks due to constrained context windows and static policies that fail to adapt to dynamic environments. To address these limitations, this work proposes the Self-Evolving GUI Agent (SE-GA), a novel framework that integrates hierarchical memory structures with an iterative self-improvement mechanism. At the core of our approach is Test-Time Memory Extension (TTME), which facilitates long-term planning by dynamically retrieving episodic, semantic, and experiential memories to provide salient contexts during inference. To ensure continuous learning, we introduce Memory-Augmented Self-Evolution (MASE), which is a training pipeline that adopts the data collected by TTME to stabilize and enhance the agent's foundational policy. Extensive evaluations across both offline and online benchmarks demonstrate SE-GA achieves state-of-the-art performance, reaching success rates of 89.0\% on ScreenSpot and 75.8\% on the challenging AndroidControl-High dataset. Furthermore, significant improvements on the AndroidWorld benchmark highlight the superior generalization to dynamic environments.

Precipitation nowcasting holds critical scientific significance and practical value for mitigating the impacts of increasingly frequent extreme weather events. Although significant progress has been made in current prediction models based on encoder decoder architecture, there are still limitations such as excessive focus on multi-scale differences and neglect of feature representation consistency, and pixel level loss functions that only focus on single point differences and ignore spatial correlation. To address these challenges, this paper proposes Clip4Rain, a novel precipitation nowcasting framework that leverages vision-language pre-trained knowledge to enable spatiotemporal feature alignment. Specifically, Clip4Rain incorporates cross-modal alignment to guide multi-scale spatiotemporal representation learning and refines precipitation pattern modeling via multi-level fusion, thereby enhancing the encoder-decoder architecture’s ability to capture complex precipitation dynamics. Comprehensive experiments demonstrate the framework’s effectiveness, achieving performance improvements of 15.35% and 15.28% in critical evaluation metrics on the KNMI dataset. On the SEVIR benchmark, it also yields substantial gains, with 6.2% and 6.4% improvements in CSI and POD, respectively.

Text-to-Image (T2I) models typically deploy safety mechanisms to prevent the generation of sensitive images. Unfortunately, recent jailbreaking attack methods manually design instructions for the LLM to generate adversarial prompts, which effectively exposing safety vulnerabilities of T2I models. However, existing methods have two limitations: 1) relying on manually exhaustive strategies for designing adversarial prompts, lacking a unified framework, and 2) requiring numerous queries to achieve a successful attack, limiting their practical applicability. To address this issue, we propose Reason2Attack~(R2A), which aims to enhance the effectiveness and efficiency of the LLM in jailbreaking attacks. Specifically, we first use Frame Semantics theory to systematize existing manually crafted strategies and propose a unified generation framework to generate CoT adversarial prompts step by step. Following this, we propose a two-stage LLM reasoning training framework guided by the attack process. In the first stage, the LLM is fine-tuned with CoT examples generated by the unified generation framework to internalize the adversarial prompt generation process grounded in Frame Semantics. In the second stage, we incorporate the jailbreaking task into the LLM's reinforcement learning process, guided by the proposed attack process reward function that balances prompt stealthiness, effectiveness, and length, enabling the LLM to understand T2I models and safety mechanisms. Extensive experiments on various T2I models with safety mechanisms, and commercial T2I models, show the superiority and practicality of R2A.

Graph anomaly detection aims to identify irregular patterns in graph-structured data. Most unsupervised GNN-based methods rely on the homophily assumption that connected nodes share similar attributes. However, real-world graphs often exhibit attribute-level heterophily, where connected nodes have dissimilar attributes. Our analysis of attribute-level heterophily graphs reveals two phenomena indicating that current approaches are not practical for unsupervised graph anomaly detection: 1) attribute similarities between connected nodes show nearly identical distributions across different connected node pair types, and 2) anomalies cause consistent variation trends between the graph with and without anomalous edges in the low- and high-frequency components of the spectral energy distributions, while the mid-part exhibits more erratic variations. Based on these observations, we propose NK-GAD, a neighbor knowledge-enhanced unsupervised graph anomaly detection framework. NK-GAD integrates a joint encoder capturing both similar and dissimilar neighbor features, a neighbor reconstruction module modeling normal distributions, a center aggregation module refining node features, and dual decoders for reconstructing attributes and structures. Experiments on seven datasets show NK-GAD achieves an average 3.29% AUC improvement.