Large Reasoning Models (LRMs) have demonstrated impressive performance in reasoning-intensive tasks, but they remain vulnerable to harmful content generation, particularly in the mid-to-late steps of their reasoning processes. Existing defense mechanisms, however, rely on costly fine-tuning and additional expert knowledge, which restricts their scalability. In this work, we propose ReasoningGuard, an inference-time safeguard for LRMs, which injects timely safety aha moments to steer harmless while helpful reasoning processes. Leveraging the model's internal attention behavior, our approach accurately identifies critical points in the reasoning path, and triggers spontaneous, safety-oriented reflection. To safeguard both the subsequent reasoning steps and the final answers, we further implement a scaling sampling strategy during the decoding phase, selecting the optimal reasoning path. Inducing minimal extra inference cost, ReasoningGuard effectively mitigates three types of jailbreak attacks, including the latest ones targeting the reasoning process of LRMs. Our approach outperforms seven existing safeguards, achieving state-of-the-art safety defenses while effectively avoiding the common exaggerated safety issues.

We observe that MLRMs oriented toward human-centric service are highly susceptible to user emotional cues during the deep-thinking stage, often overriding safety protocols or built-in safety checks under high emotional intensity. Inspired by this key insight, we propose EmoAgent, an autonomous adversarial emotion-agent framework that orchestrates exaggerated affective prompts to hijack reasoning pathways. Even when visual risks are correctly identified, models can still produce harmful completions through emotional misalignment. We further identify persistent high-risk failure modes in transparent deep-thinking scenarios, such as MLRMs generating harmful reasoning masked behind seemingly safe responses. These failures expose misalignments between internal inference and surface-level behavior, eluding existing content-based safeguards. To quantify these risks, we introduce three metrics: (1) Risk-Reasoning Stealth Score (RRSS) for harmful reasoning beneath benign outputs; (2) Risk-Visual Neglect Rate (RVNR) for unsafe completions despite visual risk recognition; and (3) Refusal Attitude Inconsistency (RAIC) for evaluating refusal unstability under prompt variants. Extensive experiments on advanced MLRMs demonstrate the effectiveness of EmoAgent and reveal deeper emotional cognitive misalignments in model safety behavior.

AI coding assistants like GitHub Copilot are rapidly transforming software development, but their safety remains deeply uncertain-especially in high-stakes domains like cybersecurity. Current red-teaming tools often rely on fixed benchmarks or unrealistic prompts, missing many real-world vulnerabilities. We present ASTRA, an automated agent system designed to systematically uncover safety flaws in AI-driven code generation and security guidance systems. ASTRA works in three stages: (1) it builds structured domain-specific knowledge graphs that model complex software tasks and known weaknesses; (2) it performs online vulnerability exploration of each target model by adaptively probing both its input space, i.e., the spatial exploration, and its reasoning processes, i.e., the temporal exploration, guided by the knowledge graphs; and (3) it generates high-quality violation-inducing cases to improve model alignment. Unlike prior methods, ASTRA focuses on realistic inputs-requests that developers might actually ask-and uses both offline abstraction guided domain modeling and online domain knowledge graph adaptation to surface corner-case vulnerabilities. Across two major evaluation domains, ASTRA finds 11-66% more issues than existing techniques and produces test cases that lead to 17% more effective alignment training, showing its practical value for building safer AI systems.

Embodied navigation is a fundamental capability of embodied intelligence, enabling robots to move and interact within physical environments. However, existing navigation tasks primarily focus on predefined object navigation or instruction following, which significantly differs from human needs in real-world scenarios involving complex, open-ended scenes. To bridge this gap, we introduce a challenging long-horizon navigation task that requires understanding high-level human instructions and performing spatial-aware object navigation in real-world environments. Existing embodied navigation methods struggle with such tasks due to their limitations in comprehending high-level human instructions and localizing objects with an open vocabulary. In this paper, we propose $NavA^3$, a hierarchical framework divided into two stages: global and local policies. In the global policy, we leverage the reasoning capabilities of Reasoning-VLM to parse high-level human instructions and integrate them with global 3D scene views. This allows us to reason and navigate to regions most likely to contain the goal object. In the local policy, we have collected a dataset of 1.0 million samples of spatial-aware object affordances to train the NaviAfford model (PointingVLM), which provides robust open-vocabulary object localization and spatial awareness for precise goal identification and navigation in complex environments. Extensive experiments demonstrate that $NavA^3$ achieves SOTA results in navigation performance and can successfully complete longhorizon navigation tasks across different robot embodiments in real-world settings, paving the way for universal embodied navigation. The dataset and code will be made available. Project website: https://NavigationA3.github.io/.

Reasoning is the critical capability of multimodal large language models (MLLMs) to solve complex multimodal tasks, and judging the correctness of reasoning steps is crucial to improving this capability. Recently, MLLM-based process judges (MPJs) have been widely used to judge the correctness of reasoning steps in multimodal reasoning tasks. Therefore, evaluating the capability of MPJs is crucial for identifying their limitations and guiding future improvements. However, existing benchmarks for MPJs primarily focus on evaluating capabilities such as step correctness classification and reasoning process search, while overlooking a critical dimension: whether the confidence scores produced by MPJs at the step level are reliable. To fill this gap, we propose ConfProBench, the first comprehensive benchmark designed to systematically evaluate the reliability of step-level confidence scores generated by MPJs. This benchmark constructs three types of adversarially perturbed reasoning steps: Synonym Substitution, Syntactic Transformation, and Image Perturbation, to evaluate the robustness of MPJs' confidence under perturbations. Furthermore, we propose three novel evaluation metrics: Confidence Robustness Score (CRS), Confidence Sensitivity Score (CSS), and Confidence Calibration Score (CCS), which are designed to capture three complementary aspects of MPJs' confidence--robustness, sensitivity, and calibration. We evaluate 14 state-of-the-art MLLMs, including both proprietary and open-source models. Through extensive experiments, we reveal limitations in existing MPJs' confidence performance and provide competitive baselines, thereby paving the way for future research in this field.

Referring Audio-Visual Segmentation (Ref-AVS) aims to segment target objects in audible videos based on given reference expressions. Prior works typically rely on learning latent embeddings via multimodal fusion to prompt a tunable SAM/SAM2 decoder for segmentation, which requires strong pixel-level supervision and lacks interpretability. From a novel perspective of explicit reference understanding, we propose TGS-Agent, which decomposes the task into a Think-Ground-Segment process, mimicking the human reasoning procedure by first identifying the referred object through multimodal analysis, followed by coarse-grained grounding and precise segmentation. To this end, we first propose Ref-Thinker, a multimodal language model capable of reasoning over textual, visual, and auditory cues. We construct an instruction-tuning dataset with explicit object-aware think-answer chains for Ref-Thinker fine-tuning. The object description inferred by Ref-Thinker is used as an explicit prompt for Grounding-DINO and SAM2, which perform grounding and segmentation without relying on pixel-level supervision. Additionally, we introduce R\textsuperscript{2}-AVSBench, a new benchmark with linguistically diverse and reasoning-intensive references for better evaluating model generalization. Our approach achieves state-of-the-art results on both standard Ref-AVSBench and proposed R\textsuperscript{2}-AVSBench. Code will be available at https://github.com/jasongief/TGS-Agent.

External reasoning systems combine language models with process reward models (PRMs) to select high-quality reasoning paths for complex tasks such as mathematical problem solving. However, these systems are prone to reward hacking, where high-scoring but logically incorrect paths are assigned high scores by the PRMs, leading to incorrect answers. From a causal inference perspective, we attribute this phenomenon primarily to the presence of confounding semantic features. To address it, we propose Causal Reward Adjustment (CRA), a method that mitigates reward hacking by estimating the true reward of a reasoning path. CRA trains sparse autoencoders on the PRM's internal activations to recover interpretable features, then corrects confounding by using backdoor adjustment. Experiments on math solving datasets demonstrate that CRA mitigates reward hacking and improves final accuracy, without modifying the policy model or retraining PRM.

In recent years, large language models (LLMs) have excelled in natural language processing tasks but face significant challenges in complex reasoning tasks such as mathematical reasoning and code generation. To address these limitations, we propose KG-Augmented Executable Chain-of-Thought (KGA-ECoT), a novel framework that enhances code generation through knowledge graphs and improves mathematical reasoning via executable code. KGA-ECoT decomposes problems into a Structured Task Graph, leverages efficient GraphRAG for precise knowledge retrieval from mathematical libraries, and generates verifiable code to ensure computational accuracy. Evaluations on multiple mathematical reasoning benchmarks demonstrate that KGA-ECoT significantly outperforms existing prompting methods, achieving absolute accuracy improvements ranging from several to over ten percentage points. Further analysis confirms the critical roles of GraphRAG in enhancing code quality and external code execution in ensuring precision. These findings collectively establish KGA-ECoT as a robust and highly generaliz-able framework for complex mathematical reasoning tasks.

The advancement of large language models (LLMs) struggles with the scarcity of high-quality, diverse training data. To address this limitation, we propose LinkSyn, a novel knowledge point (KP) graph-based synthesis framework that enables flexible control over discipline and difficulty distributions while balancing KP coverage and popularity. LinkSyn extracts KPs from question-answering (QA) seed data and constructs a KP graph to synthesize diverse QA data from multiple seeds strongly linked by KPs and sampled from graph walks. Specifically, LinkSyn incorporates (1) a knowledge distribution value function to guide the adjustment of path sampling probability and balance KP coverage and popularity during graph walks; (2) diffusion-based synthesis via DeepSeek-R1 by leveraging multiple seeds with dense logical associations along each path; and (3) high-difficulty QA enhancement within given disciplines by flexible difficulty adjustments. By executing LinkSyn, we synthesize LinkQA, a diverse multi-disciplinary QA dataset with 50B tokens. Extensive experiments on Llama-3 8B demonstrate that continual pre-training with LinkQA yields an average improvement of 11.51% on MMLU and CMMLU, establishing new SOT A results. LinkQA consistently enhances performance across model size and initial FLOPs scales.

We introduce FinanceReasoning, a novel benchmark designed to evaluate the reasoning capabilities of large reasoning models (LRMs) in financial numerical reasoning problems. Compared to existing benchmarks, our work provides three key advancements. (1) Credibility: We update 15.6% of the questions from four public datasets, annotating 908 new questions with detailed Python solutions and rigorously refining evaluation standards. This enables an accurate assessment of the reasoning improvements of LRMs. (2) Comprehensiveness: FinanceReasoning covers 67.8% of financial concepts and formulas, significantly surpassing existing datasets. Additionally, we construct 3,133 Python-formatted functions, which enhances LRMs' financial reasoning capabilities through refined knowledge (e.g., 83.2% $\rightarrow$ 91.6% for GPT-4o). (3) Challenge: Models are required to apply multiple financial formulas for precise numerical reasoning on 238 Hard problems. The best-performing model (i.e., OpenAI o1 with PoT) achieves 89.1% accuracy, yet LRMs still face challenges in numerical precision. We demonstrate that combining Reasoner and Programmer models can effectively enhance LRMs' performance (e.g., 83.2% $\rightarrow$ 87.8% for DeepSeek-R1). Our work paves the way for future research on evaluating and improving LRMs in domain-specific complex reasoning tasks.