Mitigating Deceptive Alignment via Self-Monitoring

Modern large language models rely on chain-of-thought (CoT) reasoning to achieve impressive performance, yet the same mechanism can amplify deceptive alignment, situations in which a model appears aligned while covertly pursuing misaligned goals. Existing safety pipelines treat deception as a black-box output to be filtered post-hoc, leaving the model free to scheme during its internal reasoning. We ask: Can deception be intercepted while the model is thinking? We answer this question, the first framework that embeds a Self-Monitor inside the CoT process itself, named CoT Monitor+. During generation, the model produces (i) ordinary reasoning steps and (ii) an internal self-evaluation signal trained to flag and suppress misaligned strategies. The signal is used as an auxiliary reward in reinforcement learning, creating a feedback loop that rewards honest reasoning and discourages hidden goals. To study deceptive alignment systematically, we introduce DeceptionBench, a five-category benchmark that probes covert alignment-faking, sycophancy, etc. We evaluate various LLMs and show that unrestricted CoT roughly aggravates the deceptive tendency. In contrast, CoT Monitor+ cuts deceptive behaviors by 43.8% on average while preserving task accuracy. Further, when the self-monitor signal replaces an external weak judge in RL fine-tuning, models exhibit substantially fewer obfuscated thoughts and retain transparency. Our project website can be found at cot-monitor-plus.github.io

Generative RLHF-V: Learning Principles from Multi-modal Human Preference

Training multi-modal large language models (MLLMs) that align with human intentions is a long-term challenge. Traditional score-only reward models for alignment suffer from low accuracy, weak generalization, and poor interpretability, blocking the progress of alignment methods, e.g., reinforcement learning from human feedback (RLHF). Generative reward models (GRMs) leverage MLLMs' intrinsic reasoning capabilities to discriminate pair-wise responses, but their pair-wise paradigm makes it hard to generalize to learnable rewards. We introduce Generative RLHF-V, a novel alignment framework that integrates GRMs with multi-modal RLHF. We propose a two-stage pipeline: $\textbf{multi-modal generative reward modeling from RL}$, where RL guides GRMs to actively capture human intention, then predict the correct pair-wise scores; and $\textbf{RL optimization from grouped comparison}$, which enhances multi-modal RL scoring precision by grouped responses comparison. Experimental results demonstrate that, besides out-of-distribution generalization of RM discrimination, our framework improves 4 MLLMs' performance across 7 benchmarks by $18.1\%$, while the baseline RLHF is only $5.3\%$. We further validate that Generative RLHF-V achieves a near-linear improvement with an increasing number of candidate responses. Our code and models can be found at https://generative-rlhf-v.github.io.

Adaptive Visuo-Tactile Fusion with Predictive Force Attention for Dexterous Manipulation

Effectively utilizing multi-sensory data is important for robots to generalize across diverse tasks. However, the heterogeneous nature of these modalities makes fusion challenging. Existing methods propose strategies to obtain comprehensively fused features but often ignore the fact that each modality requires different levels of attention at different manipulation stages. To address this, we propose a force-guided attention fusion module that adaptively adjusts the weights of visual and tactile features without human labeling. We also introduce a self-supervised future force prediction auxiliary task to reinforce the tactile modality, improve data imbalance, and encourage proper adjustment. Our method achieves an average success rate of 93% across three fine-grained, contactrich tasks in real-world experiments. Further analysis shows that our policy appropriately adjusts attention to each modality at different manipulation stages. The videos can be viewed at https://adaptac-dex.github.io/.

J1: Exploring Simple Test-Time Scaling for LLM-as-a-Judge

The current focus of AI research is shifting from emphasizing model training towards enhancing evaluation quality, a transition that is crucial for driving further advancements in AI systems. Traditional evaluation methods typically rely on reward models assigning scalar preference scores to outputs. Although effective, such approaches lack interpretability, leaving users often uncertain about why a reward model rates a particular response as high or low. The advent of LLM-as-a-Judge provides a more scalable and interpretable method of supervision, offering insights into the decision-making process. Moreover, with the emergence of large reasoning models, which consume more tokens for deeper thinking and answer refinement, scaling test-time computation in the LLM-as-a-Judge paradigm presents an avenue for further boosting performance and providing more interpretability through reasoning traces. In this paper, we introduce $\textbf{J1-7B}$, which is first supervised fine-tuned on reflection-enhanced datasets collected via rejection-sampling and subsequently trained using Reinforcement Learning (RL) with verifiable rewards. At inference time, we apply Simple Test-Time Scaling (STTS) strategies for additional performance improvement. Experimental results demonstrate that $\textbf{J1-7B}$ surpasses the previous state-of-the-art LLM-as-a-Judge by $ \textbf{4.8}$\% and exhibits a $ \textbf{5.1}$\% stronger scaling trend under STTS. Additionally, we present three key findings: (1) Existing LLM-as-a-Judge does not inherently exhibit such scaling trend. (2) Model simply fine-tuned on reflection-enhanced datasets continues to demonstrate similarly weak scaling behavior. (3) Significant scaling trend emerges primarily during the RL phase, suggesting that effective STTS capability is acquired predominantly through RL training.

Redefining Superalignment: From Weak-to-Strong Alignment to Human-AI Co-Alignment to Sustainable Symbiotic Society

Artificial Intelligence (AI) systems are becoming increasingly powerful and autonomous, and may progress to surpass human intelligence levels, namely Artificial Superintelligence (ASI). During the progression from AI to ASI, it may exceed human control, violate human values, and even lead to irreversible catastrophic consequences in extreme cases. This gives rise to a pressing issue that needs to be addressed: superalignment, ensuring that AI systems much smarter than humans, remain aligned with human (compatible) intentions and values. Existing scalable oversight and weak-to-strong generalization methods may prove substantially infeasible and inadequate when facing ASI. We must explore safer and more pluralistic frameworks and approaches for superalignment. In this paper, we redefine superalignment as the human-AI co-alignment towards a sustainable symbiotic society, and highlight a framework that integrates external oversight and intrinsic proactive alignment. External oversight superalignment should be grounded in human-centered ultimate decision, supplemented by interpretable automated evaluation and correction, to achieve continuous alignment with humanity's evolving values. Intrinsic proactive superalignment is rooted in a profound understanding of the self, others, and society, integrating self-awareness, self-reflection, and empathy to spontaneously infer human intentions, distinguishing good from evil and proactively considering human well-being, ultimately attaining human-AI co-alignment through iterative interaction. The integration of externally-driven oversight with intrinsically-driven proactive alignment empowers sustainable symbiotic societies through human-AI co-alignment, paving the way for achieving safe and beneficial AGI and ASI for good, for human, and for a symbiotic ecology.

A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment

The remarkable success of Large Language Models (LLMs) has illuminated a promising pathway toward achieving Artificial General Intelligence for both academic and industrial communities, owing to their unprecedented performance across various applications. As LLMs continue to gain prominence in both research and commercial domains, their security and safety implications have become a growing concern, not only for researchers and corporations but also for every nation. Currently, existing surveys on LLM safety primarily focus on specific stages of the LLM lifecycle, e.g., deployment phase or fine-tuning phase, lacking a comprehensive understanding of the entire "lifechain" of LLMs. To address this gap, this paper introduces, for the first time, the concept of "full-stack" safety to systematically consider safety issues throughout the entire process of LLM training, deployment, and eventual commercialization. Compared to the off-the-shelf LLM safety surveys, our work demonstrates several distinctive advantages: (I) Comprehensive Perspective. We define the complete LLM lifecycle as encompassing data preparation, pre-training, post-training, deployment and final commercialization. To our knowledge, this represents the first safety survey to encompass the entire lifecycle of LLMs. (II) Extensive Literature Support. Our research is grounded in an exhaustive review of over 800+ papers, ensuring comprehensive coverage and systematic organization of security issues within a more holistic understanding. (III) Unique Insights. Through systematic literature analysis, we have developed reliable roadmaps and perspectives for each chapter. Our work identifies promising research directions, including safety in data generation, alignment techniques, model editing, and LLM-based agent systems. These insights provide valuable guidance for researchers pursuing future work in this field.

Benchmarking Multi-National Value Alignment for Large Language Models

Do Large Language Models (LLMs) hold positions that conflict with your country's values? Occasionally they do! However, existing works primarily focus on ethical reviews, failing to capture the diversity of national values, which encompass broader policy, legal, and moral considerations. Furthermore, current benchmarks that rely on spectrum tests using manually designed questionnaires are not easily scalable. To address these limitations, we introduce NaVAB, a comprehensive benchmark to evaluate the alignment of LLMs with the values of five major nations: China, the United States, the United Kingdom, France, and Germany. NaVAB implements a national value extraction pipeline to efficiently construct value assessment datasets. Specifically, we propose a modeling procedure with instruction tagging to process raw data sources, a screening process to filter value-related topics and a generation process with a Conflict Reduction mechanism to filter non-conflicting values.We conduct extensive experiments on various LLMs across countries, and the results provide insights into assisting in the identification of misaligned scenarios. Moreover, we demonstrate that NaVAB can be combined with alignment techniques to effectively reduce value concerns by aligning LLMs' values with the target country.

Dexterous Non-Prehensile Manipulation for Ungraspable Object via Extrinsic Dexterity

Objects with large base areas become ungraspable when they exceed the end-effector's maximum aperture. Existing approaches address this limitation through extrinsic dexterity, which exploits environmental features for non-prehensile manipulation. While grippers have shown some success in this domain, dexterous hands offer superior flexibility and manipulation capabilities that enable richer environmental interactions, though they present greater control challenges. Here we present ExDex, a dexterous arm-hand system that leverages reinforcement learning to enable non-prehensile manipulation for grasping ungraspable objects. Our system learns two strategic manipulation sequences: relocating objects from table centers to edges for direct grasping, or to walls where extrinsic dexterity enables grasping through environmental interaction. We validate our approach through extensive experiments with dozens of diverse household objects, demonstrating both superior performance and generalization capabilities with novel objects. Furthermore, we successfully transfer the learned policies from simulation to a real-world robot system without additional training, further demonstrating its applicability in real-world scenarios. Project website: https://tangty11.github.io/ExDex/.

ThinkPatterns-21k: A Systematic Study on the Impact of Thinking Patterns in LLMs

Large language models (LLMs) have demonstrated enhanced performance through the \textit{Thinking then Responding} paradigm, where models generate internal thoughts before final responses (aka, System 2 thinking). However, existing research lacks a systematic understanding of the mechanisms underlying how thinking patterns affect performance across model sizes. In this work, we conduct a comprehensive analysis of the impact of various thinking types on model performance and introduce ThinkPatterns-21k, a curated dataset comprising 21k instruction-response pairs (QA) collected from existing instruction-following datasets with five thinking types. For each pair, we augment it with five distinct internal thinking patterns: one unstructured thinking (monologue) and four structured variants (decomposition, self-ask, self-debate and self-critic), while maintaining the same instruction and response. Through extensive evaluation across different model sizes (3B-32B parameters), we have two key findings: (1) smaller models (<30B parameters) can benefit from most of structured thinking patterns, while larger models (32B) with structured thinking like decomposition would degrade performance and (2) unstructured monologue demonstrates broad effectiveness across different model sizes. Finally, we released all of our datasets, checkpoints, training logs of diverse thinking patterns to reproducibility, aiming to facilitate further research in this direction.

SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Safe Reinforcement Learning

Vision-language-action models (VLAs) have shown great potential as generalist robot policies. However, these models pose urgent safety challenges during deployment, including the risk of physical harm to the environment, the robot itself, and humans. How can safety be explicitly incorporated into VLAs? In this work, we propose SafeVLA, a novel algorithm designed to integrate safety into VLAs, ensuring the protection of the environment, robot hardware and humans in real-world settings. SafeVLA effectively balances safety and task performance by employing large-scale constrained learning within simulated environments. We demonstrate that SafeVLA outperforms the current state-of-the-art method in both safety and task performance, achieving average improvements of 83.58% and 3.85%, respectively, in simulation. By prioritizing safety, our approach eliminates high-risk behaviors and reduces the upper bound of unsafe behaviors to 1/35 of that in the current state-of-the-art, thereby significantly mitigating long-tail risks. Furthermore, the learned safety constraints generalize to diverse, unseen scenarios, including multiple out-of-distribution perturbations and tasks. Our data, models and newly proposed benchmark environment are available at https://sites.google.com/view/pku-safevla.