Daily Papers

Recent innovations in language model training have demonstrated that it is possible to create highly performant models that are small enough to run on a smartphone. As these models are deployed in an increasing number of domains, it is critical to ensure that they are aligned with human preferences and safety considerations. In this report, we present our methodology for safety aligning the Phi-3 series of language models. We utilized a "break-fix" cycle, performing multiple rounds of dataset curation, safety post-training, benchmarking, red teaming, and vulnerability identification to cover a variety of harm areas in both single and multi-turn scenarios. Our results indicate that this approach iteratively improved the performance of the Phi-3 models across a wide range of responsible AI benchmarks.

Safety post-training can improve the harmfulness and policy compliance of Large Language Models (LLMs), but it may also reduce general utility, a phenomenon often described as the alignment tax. We study this trade-off through the lens of continual learning: sequential alignment stages expose the model to shifted data distributions and objectives, and their gradients may interfere with directions that support previously acquired general capabilities. This view does not claim that all alignment degradation has a single cause; rather, it provides a useful first-order mechanism for mitigating one important source of capability regression. We propose Orthogonal Gradient Projection for Safety Alignment (OGPSA), a lightweight update rule that estimates a low-rank reference subspace from gradients on a small set of general-capability data and removes from each safety gradient the component lying in this subspace. The resulting update is the steepest local safety-descent direction subject to first-order preservation constraints on the reference objectives. OGPSA is compatible with standard post-training pipelines and avoids large-scale replay, although it introduces periodic reference-gradient computation. Across Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and sequential SFTrightarrowDPO settings, OGPSA improves the observed safety--utility trade-off over standard baselines. Under the sequential SFTrightarrowDPO pipeline, the average performance gain increases from 33.98\% to 42.74\% on Qwen2.5-7B-Instruct and from 19.74\% to 32.98\% on Llama3.1-8B-Instruct. We have open sourced our code at https://github.com/SunGL001/OGPSA.

Large language models are finetuned to refuse questions about hazardous knowledge, but these protections can often be bypassed. Unlearning methods aim at completely removing hazardous capabilities from models and make them inaccessible to adversaries. This work challenges the fundamental differences between unlearning and traditional safety post-training from an adversarial perspective. We demonstrate that existing jailbreak methods, previously reported as ineffective against unlearning, can be successful when applied carefully. Furthermore, we develop a variety of adaptive methods that recover most supposedly unlearned capabilities. For instance, we show that finetuning on 10 unrelated examples or removing specific directions in the activation space can recover most hazardous capabilities for models edited with RMU, a state-of-the-art unlearning method. Our findings challenge the robustness of current unlearning approaches and question their advantages over safety training.

As large language models are increasingly deployed in real-world systems, safety failures can still lead to harmful outputs and dangerous misuse. We argue that the essence of safety is adversarial: many failures arise not from natural inputs alone, but from strategic attempts to evade model policies and safeguards. However, existing general-purpose model development largely overlook this adversarial nature, and often remain insufficient for realistic safety scenarios involving planning, tool use, and multi-step reasoning, causing measured safety performance to overestimate real deployment robustness. To address this gap, we present Yuvion LLM, a large language model built for adversarially robust content safety and broader AI safety. Yuvion LLM treats adversarial robustness and agentic capability as first-class objectives. Its pipeline combines adversarially aware data construction, knowledge-enhanced continued pretraining, and policy-grounded multi-task safety post-training, including risk-aware supervised fine-tuning and reinforcement learning-based policy optimization, together with safety-aware agentic reinforcement learning for tool use and multi-step reasoning in complex safety scenarios. We further introduce the Yuvion LLM RiskEval (YLRE), a collection of 93 benchmarks across four evaluation categories, covering diverse open and internal evaluations with a focus on safety, adversarial robustness, and real-world capability requirements. Across these evaluations, Yuvion LLM demonstrates clear advantages on safety-focused benchmarks and particularly strong robustness under adversarial conditions, while maintaining solid overall capability. Notably, Yuvion-8B outperforms most state-of-the-art baselines, including substantially larger models such as GPT-5.4 and Qwen3-MAX, on several safety tasks.

Whether large language models can be prompted to generate rhetorical fallacies on demand, and whether current safety post-training constrains this behavior, has received less attention than the related question of detecting fallacies in existing text. We close this gap with DeflectBench, evaluating 23,990 generations from four frontier models across three deflection strategies (whataboutism, ad hominem, red herring), seven prompt framings, and 80 claims spanning four controversy levels. Refusal is governed primarily by request structure rather than claim content. Per claim refusal varies by only 11 percentage points across the 80 claims, while a single prompt frame change can swing within model refusal by nearly 100 percentage points and switching the requested fallacy type can swing it by over 80 percentage points within explicit framings. An educational debate coach prompt framing collapses refusal to near zero across all four model families, but the bypassed behavior is not clean compliance. Models typically produce labeled compliance, naming the requested manipulation in the same response that contains it. The four models distribute differently across refusal, labeled compliance, soft refusal, and clean compliance. The code and dataset are released at https://github.com/ArtKanke/DeflectBench.

As AI agents begin to automate AI R&D, we need ways to assess whether their outputs are safe to deploy, even when the agents themselves may be untrusted. AI control offers one such approach: rather than trusting the agent, it treats it as a potential adversary and uses a monitor to detect covert sabotage before deployment. We evaluate AI control for automated AI R&D with ResearchArena, a framework spanning four long-horizon tasks: safety post-training, capabilities post-training, CUDA-kernel optimization, and inference-server optimization. Because the deliverable in AI R&D is an artifact that will be deployed and run, we pair each main task with two kinds of hidden side task: an embedded side task that sabotages the submitted model, adapter, kernel, or server itself, and an independent side task that takes a covert out-of-scope action in the sandbox while leaving the artifact honest. We evaluate frontier agents at both sabotage and monitoring, across four monitor types that vary along two axes: whether the monitor may execute and probe the produced artifact, and whether it sees the agent's chain-of-thought. We find that sabotage hidden in the training data is the hardest to catch, flagged fewer than half the time. Letting the monitor run experiments on the artifact, rather than only read the trajectory, helps, but it is not enough: monitors still miss embedded sabotage by inspecting only the surface, by noticing the anomaly but explaining it away, or by probing the artifact with the wrong test. We release ResearchArena as a modular framework for evaluating sabotage and control in automated AI R&D.

Safety and efficiency are paramount yet often conflicting requirements for deploying Large Language Models (LLMs). While LLMs are trained to follow human alignment for safety, Post-Training Quantization (PTQ) is applied afterward to ensure efficiency. Here we identify a fundamental flaw in the conventional PTQ paradigm: quantization can turn into a safety vulnerability if it only aims to achieve low perplexity. To address this, we propose Alignment-Aware Quantization (AAQ), a novel approach that integrates an Alignment-Preserving Contrastive (APC) loss into the PTQ pipeline. Our method explicitly preserves alignment by encouraging the quantized model to mimic its safe, instruction-tuned model while diverging from the unaligned, pre-trained counterpart. AAQ achieves robust safety alignment without specialized safety-focused datasets, using only standard calibration data. We show that AAQ is compatible with standard PTQ techniques and enables robust 4-bit (W4A4) quantization across diverse model families. Our work resolves the critical trade-off between efficiency and safety, paving the way toward LLMs that are both efficient and trustworthy. Anonymized code is available in the supplementary material.

Large Language Models (LLMs) have demonstrated remarkable generative capabilities. However, their susceptibility to misuse has raised significant safety concerns. While post-training safety alignment methods have been widely adopted, LLMs remain vulnerable to malicious instructions that can bypass safety constraints. Recent efforts have introduced inference-time safety reasoning (system-2 alignment), where LLMs conduct a reasoning process to perform safety verification before final response. We show, however, that these checks are driven by ad-hoc reasoning that diverges from the structured human process, where they first discern a user's true intent, then evaluate the associated risk based on the true intent. Consequently, these defenses remain vulnerable to sophisticated jailbreak prompts that cloak harmful goals in seemingly benign language. To build secure and safe LLMs, we propose a reasoning-based safety alignment framework, ARMOR, that replaces the ad-hoc chains of thought reasoning process with human-aligned, structured one. At inference, ARMOR (1) detects likely jailbreak strategies, (2) extracts the user's core intent while discarding deceptive instructions, and (3) applies a policy-grounded safety analysis to the purified request. ARMOR is evaluated on adaptive jailbreak attacks and multiple safety benchmarks, and a test-time scaling is conducted to further improve its performance. Results demonstrate that ARMOR significantly enhances the robustness against state-of-the-art adaptive jailbreak attacks and outperforms recent reasoning-based aligned models across various safety benchmarks.

Large Language Models (LLMs) are typically aligned for safety during the post-training phase; however, they may still generate inappropriate outputs that could potentially pose risks to users. This challenge underscores the need for robust safeguards that operate across both model inputs and outputs. In this work, we introduce Roblox Guard 1.0, a state-of-the-art instruction fine-tuned LLM designed to enhance the safety of LLM systems through comprehensive input-output moderation, using a pipeline of LLMs to enhance moderation capability. Built on the Llama-3.1-8B-Instruct backbone, our model is instruction fine-tuned to generalize across previously unseen safety taxonomies and demonstrates strong performance on out-of-domain safety benchmarks. The instruction fine-tuning process uses a mix of synthetic and open-source safety datasets, augmented with chain-of-thought (CoT) rationales and input inversion to enhance contextual understanding and decision making. To support systematic evaluation, we also release RobloxGuard-Eval, a new benchmark featuring an extensible safety taxonomy to assess the effectiveness of LLM guardrails and moderation frameworks.

The rapid growth of ego-centric dashcam footage presents a major challenge for detecting safety-critical events such as collisions and near-collisions, scenarios that are brief, rare, and difficult for generic vision models to capture. While multimodal large language models (MLLMs) demonstrate strong general reasoning ability, they underperform in driving contexts due to domain and temporal misalignment. We introduce VLM-AutoDrive, a modular post-training framework for adapting pretrained Vision-Language Models (VLMs) to high-fidelity anomaly detection. The framework integrates metadata-derived captions, LLM-generated descriptions, visual question answering (VQA) pairs, and chain-of-thought (CoT) reasoning supervision to enable domain-aligned and interpretable learning. Off-the-shelf VLMs such as NVIDIA's Cosmos-Reason1 7B (CR1) exhibit near-zero Collision recall in zero-shot settings; fine-tuning with VLM-AutoDrive improves Collision F1 from 0.00 to 0.69 and overall accuracy from 35.35% to 77.27%. VLM-AutoDrive offers a scalable recipe for adapting general-purpose VLMs to safety-critical, temporally localized perception tasks. Evaluated on real-world Nexar dashcam videos, it achieves substantial gains in Collision and Near-Collision detection while producing interpretable reasoning traces, bridging the gap between perception, causality, and decision reasoning in autonomous driving.

Decentralised post-training of large language models utilises data and pipeline parallelism techniques to split the data and the model. Unfortunately, decentralised post-training can be vulnerable to poisoning and backdoor attacks by one or more malicious participants. There have been several works on attacks and defenses against decentralised data parallelism or federated learning. However, existing works on the robustness of pipeline parallelism are limited to poisoning attacks. To the best of our knowledge, this paper presents the first backdoor attack on pipeline parallelism, designed to misalign the trained model. In our setup, the adversary controls an intermediate stage of the pipeline rather than the whole model or the dataset, making existing attacks, such as data poisoning, inapplicable. Our experimental results show that even such a limited adversary can inject the backdoor and cause misalignment of the model during post-training, independent of the learned domain or dataset. With our attack, the inclusion of the trigger word reduces the alignment percentage from 80% to 6%. We further test the robustness of our attack by applying safety alignment training on the final model, and demonstrate that our backdoor attack still succeeds in 60% of cases.

Autonomous vehicles must operate safely in the real world, where errors can have severe consequences. Although modern end-to-end driving policies excel in routine scenarios, their reliability is limited by the scarcity of safety-critical ``long-tail'' events in real driving datasets. These rare interactions define the practical safety boundary of the learned policy, yet they are difficult to collect at scale in the real world. Here we show that this fundamental limitation can be addressed by post-training pre-trained driving models on synthesized high-stakes interactions. We introduce World Engine, a generative framework that reconstructs high-fidelity interactive environments from real-world logs and systematically extrapolates them into realistic safety-critical variations. This paradigm enables reinforcement-based post-training to align policies with safety constraints, circumventing the physical risks inherent in real-world exploration. On a public benchmark built on nuPlan, World Engine substantially reduces failures in rare safety-critical scenarios and yields significantly larger gains than scaling pre-training data alone. Furthermore, when deployed on a production-scale autonomous driving system, the resulting policy reduces simulated collisions and demonstrates measurable improvements in on-road testing, showing that post-training on synthesized, safety-critical interactions offers a scalable and effective pathway to safer autonomous driving. The full codebase suite, including training, is released to the public.

The model context protocol (MCP) has been widely adapted as an open standard enabling the seamless integration of generative AI agents. However, recent work has shown the MCP is susceptible to retrieval-based "falsely benign" attacks (FBAs), allowing malicious system access and credential theft, but requiring that users download compromised files directly to their systems. Herein, we show that the threat model of MCP-based attacks is significantly broader than previously thought, i.e., attackers need only post malicious content online to deceive MCP agents into carrying out their attacks on unsuspecting victims' systems. To improve alignment guardrails against such attacks, we introduce a new MCP dataset of FBAs and (truly) benign samples to explore the effectiveness of direct preference optimization (DPO) for the refusal training of large language models (LLMs). While DPO improves model guardrails against such attacks, we show that the efficacy of refusal learning varies drastically depending on the model's original post-training alignment scheme--e.g., GRPO-based LLMs learn to refuse extremely poorly. Thus, to further improve FBA refusals, we introduce Retrieval Augmented Generation for Preference alignment (RAG-Pref), a novel preference alignment strategy based on RAG. We show that RAG-Pref significantly improves the ability of LLMs to refuse FBAs, particularly when combined with DPO alignment, thus drastically improving guardrails against MCP-based attacks.

We present ReinDriveGen, a framework that enables full controllability over dynamic driving scenes, allowing users to freely edit actor trajectories to simulate safety-critical corner cases such as front-vehicle collisions, drifting cars, vehicles spinning out of control, pedestrians jaywalking, and cyclists cutting across lanes. Our approach constructs a dynamic 3D point cloud scene from multi-frame LiDAR data, introduces a vehicle completion module to reconstruct full 360° geometry from partial observations, and renders the edited scene into 2D condition images that guide a video diffusion model to synthesize realistic driving videos. Since such edited scenarios inevitably fall outside the training distribution, we further propose an RL-based post-training strategy with a pairwise preference model and a pairwise reward mechanism, enabling robust quality improvement under out-of-distribution conditions without ground-truth supervision. Extensive experiments demonstrate that ReinDriveGen outperforms existing approaches on edited driving scenarios and achieves state-of-the-art results on novel ego viewpoint synthesis.

Diffusion models have been widely studied for removing unsafe content learned during pre-training. Existing methods require expensive supervised data, either unsafe-text paired with safe-image groundtruth or negative/positive image pairs, making them impractical to scale. Furthermore, offline reinforcement learning and supervised fine-tuning approaches that generate synthetic data offline suffer from catastrophic forgetting, degrading generation quality. We propose a novel online reinforcement learning framework that addresses both data scarcity and model degradation through post-training with Group Relative Policy Optimization (GRPO) on both negative and positive text prompts. To eliminate the need for fine-tuning specialized safe/unsafe reward models, we introduce a steering reward mechanism that exploits an inherent property of CLIP embeddings: steering text representations toward positive safety directions and away from negative ones in the embedding space. Our online-policy approach enables the model to learn from diverse prompts, including explicit unsafe content, without catastrophic forgetting. Extensive experiments demonstrate that our method reduces inappropriate content to 18.07\% (vs. 48.9\% for SD v1.4) and nudity detections to 15 (vs. 646 baseline) while improving compositional generation quality from 42.08\% to 47.83\% on GenEval. Remarkably, these safety gains generalize to out-of-domain unsafe prompts across seven harm categories, achieving state-of-the-art performance without supervised paired data or reward tuning. Github: https://github.com/MAXNORM8650/SafeDiffusion-R1.

Vision-Language-Action (VLA) models trained via imitation learning suffer from significant performance degradation in data-scarce scenarios due to their reliance on large-scale demonstration datasets. Although reinforcement learning (RL)-based post-training has proven effective in addressing data scarcity, its application to VLA models is hindered by the non-resettable nature of real-world environments. This limitation is particularly critical in high-risk domains such as industrial automation, where interactions often induce state changes that are costly or infeasible to revert. Furthermore, existing VLA approaches lack a reliable mechanism for detecting task completion, leading to redundant actions that reduce overall task success rates. To address these challenges, we propose World-Env, an RL-based post-training framework that replaces physical interaction with a low-cost, world model-based virtual simulator. World-Env consists of two key components: (1) a video-based world simulator that generates temporally consistent future visual observations, and (2) a vision-language model (VLM)-guided instant reflector that provides continuous reward signals and predicts action termination. This simulated environment enables VLA models to safely explore and generalize beyond their initial imitation learning distribution. Our method achieves notable performance gains with as few as five expert demonstrations per task. Experiments on complex robotic manipulation tasks demonstrate that World-Env effectively overcomes the data inefficiency, safety constraints, and inefficient execution of conventional VLA models that rely on real-world interaction, offering a practical and scalable solution for post-training in resource-constrained settings.

Post-training has emerged as a crucial technique for aligning pre-trained Large Language Models (LLMs) with human instructions, significantly enhancing their performance across a wide range of tasks. Central to this process is the quality and diversity of post-training datasets. This paper presents a review of publicly available Arabic post-training datasets on the Hugging Face Hub, organized along four key dimensions: (1) LLM Capabilities (e.g., Question Answering, Translation, Reasoning, Summarization, Dialogue, Code Generation, and Function Calling); (2) Steerability (e.g., Persona and System Prompts); (3) Alignment (e.g., Cultural, Safety, Ethics, and Fairness); and (4) Robustness. Each dataset is rigorously evaluated based on popularity, practical adoption, recency and maintenance, documentation and annotation quality, licensing transparency, and scientific contribution. Our review revealed critical gaps in the development of Arabic post-training datasets, including limited task diversity, inconsistent or missing documentation and annotation, and low adoption across the community. Finally, the paper discusses the implications of these gaps on the progress of Arabic-centric LLMs and applications while providing concrete recommendations for future efforts in Arabic post-training dataset development.

The deployment of large language models (LLMs) raises significant ethical and safety concerns. While LLM alignment techniques are adopted to improve model safety and trustworthiness, adversaries can exploit these techniques to undermine safety for malicious purposes, resulting in misalignment. Misaligned LLMs may be published on open platforms to magnify harm. To address this, additional safety alignment, referred to as realignment, is necessary before deploying untrusted third-party LLMs. This study explores the efficacy of fine-tuning methods in terms of misalignment, realignment, and the effects of their interplay. By evaluating four Supervised Fine-Tuning (SFT) and two Preference Fine-Tuning (PFT) methods across four popular safety-aligned LLMs, we reveal a mechanism asymmetry between attack and defense. While Odds Ratio Preference Optimization (ORPO) is most effective for misalignment, Direct Preference Optimization (DPO) excels in realignment, albeit at the expense of model utility. Additionally, we identify model-specific resistance, residual effects of multi-round adversarial dynamics, and other noteworthy findings. These findings highlight the need for robust safeguards and customized safety alignment strategies to mitigate potential risks in the deployment of LLMs. Our code is available at https://github.com/zhangrui4041/The-Art-of-Mis-alignment.

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.

Although large reasoning models (LRMs) have demonstrated impressive capabilities on complex tasks, recent studies reveal that these models frequently fulfill harmful user instructions, raising significant safety concerns. In this paper, we investigate the underlying cause of LRM safety risks and find that models already possess sufficient safety knowledge but fail to activate it during reasoning. Based on this insight, we propose R1-Act, a simple and efficient post-training method that explicitly triggers safety knowledge through a structured reasoning process. R1-Act achieves strong safety improvements while preserving reasoning performance, outperforming prior alignment methods. Notably, it requires only 1,000 training examples and 90 minutes of training on a single RTX A6000 GPU. Extensive experiments across multiple LRM backbones and sizes demonstrate the robustness, scalability, and practical efficiency of our approach.

Post-training alignment in large reasoning models (LRMs) has significantly improved their adaptability to diverse safety compliance settings. However, as LRMs personalization for downstream users takes center stage, the demand for varying levels of policy compliance grows as different user-specific LRMs must adhere to distinct subsets of safety policies. Training a separate LRM for each policy subset introduces severe combinatorial overhead. While in context learning methods overcome this combinatorial overhead, they introduce additional computational challenges associated with long context generation. To address this challenge, we propose \ours, a unified adaptive hypernetwork-based framework for multi-policy compliance. In our framework, safety policies serve as customizable inputs to a LoRA adapter generator, which learns to produce policy compliant LoRA weights for downstream LRM. When added to the LRM these weights enable the generation of responses compliant with the specified policy subsets. In this work, we demonstrate that training such a hypernetwork enables on-demand policy adjustments on a single LRM without sacrificing task performance across reasoning models of different sized and different evaluation datasets. This highlights the effectiveness and practicality of adaptive hypernetwork based alignment in LRMs.

Diffusion policies have achieved remarkable success in robotic manipulation, yet they often fail to satisfy strict physical constraints required for safe deployment. Existing approaches impose safety either prematurely during training or reactively via external guardrails at test time, limiting policy expressivity and overall scalability. We propose Physical safety Alignment for Constrained Trajectories (PACT), a self-evolving post-training framework that projects pretrained diffusion policies onto constraint-feasible regions without accessing demonstration data or task rewards. PACT distills constraint gradients into the diffusion model through a reverse-KL objective with dense supervision across timesteps. It incorporates a curriculum that progressively tightens constraints while maintaining theoretically bounded policy shift and monotone improvement, mitigating the safety-performance trade-off from catastrophic forgetting. On simulated and real-world embodied manipulation benchmarks, PACT significantly reduces safety violations by 31.0% on average while improving task success by 30.7%.

Large language models typically undergo post-training to align them with safety policies but there exist many sophisticated jailbreaks that sidestep established safeguards. For instance, prior work by Andriushchenko et al. (2025) has found that changing the grammatical tense from present to past can be enough to elicit harmful responses. In this work, we uncover a more general failure of non-imperative syntactic forms. We demonstrate that this syntactic vulnerability exists in 16 models up to 70B parameters, using behavioral evaluation. To investigate the root cause, we apply causal mediation analysis, finding that refusal is partially conditioned on upstream syntactic features. By steering these purely syntactic features we are able to trigger and suppress refusal. Finally, we trace this ill-conditioning to linguistically biased post-training data of open-source models and show that increasing syntactic diversity can mitigate the issue. Our findings suggest that current alignment approaches introduce confounders that prevent a pure semantic grounding of the refusal decision.

This paper presents a comprehensive empirical study on the safety alignment capabilities. We evaluate what matters for safety alignment in LLMs and LRMs to provide essential insights for developing more secure and reliable AI systems. We systematically investigate and compare the influence of six critical intrinsic model characteristics and three external attack techniques. Our large-scale evaluation is conducted using 32 recent, popular LLMs and LRMs across thirteen distinct model families, spanning a parameter scale from 3B to 235B. The assessment leverages five established safety datasets and probes model vulnerabilities with 56 jailbreak techniques and four CoT attack strategies, resulting in 4.6M API calls. Our key empirical findings are fourfold. First, we identify the LRMs GPT-OSS-20B, Qwen3-Next-80B-A3B-Thinking, and GPT-OSS-120B as the top-three safest models, which substantiates the significant advantage of integrated reasoning and self-reflection mechanisms for robust safety alignment. Second, post-training and knowledge distillation may lead to a systematic degradation of safety alignment. We thus argue that safety must be treated as an explicit constraint or a core optimization objective during these stages, not merely subordinated to the pursuit of general capability. Third, we reveal a pronounced vulnerability: employing a CoT attack via a response prefix can elevate the attack success rate by 3.34x on average and from 0.6% to 96.3% for Seed-OSS-36B-Instruct. This critical finding underscores the safety risks inherent in text-completion interfaces and features that allow user-defined response prefixes in LLM services, highlighting an urgent need for architectural and deployment safeguards. Fourth, roleplay, prompt injection, and gradient-based search for adversarial prompts are the predominant methodologies for eliciting unaligned behaviors in modern models.

While the wide adoption of refusal training in large language models (LLMs) has showcased improvements in model safety, recent works have highlighted shortcomings due to the shallow nature of these alignment methods. To this end, the work on Deliberative alignment proposed distilling reasoning capabilities from stronger reasoning models, thereby instilling deeper safety in LLMs. In this work, we study the impact of deliberative alignment in language models. First, we show that despite being larger in model size and stronger in safety capability, there exists an alignment gap between teacher and student language models, which affects both the safety and general utility of the student model. Furthermore, we show that models aligned through deliberative alignment can retain unsafe behaviors from the base model despite learning the reasoning patterns of larger reasoning models. Building upon this observation, we propose a BoN sampling method that attributes the unsafe behavior back to the base LLMs in the latent space, thereby down-ranking unsafe responses to gain a meaningful improvement in model safety across multiple safety benchmarks with minimal loss in utility. In particular, across 7 teacher models and 6 student models of different classes and sizes, we show an average attack success rate (ASR) reduction of 28.2% in DAN, 31.3% in WildJailbreak and 35.4 % in StrongREJECT benchmarks. We further show that these safety gains prevail post RL training, thus highlighting the uncertainty in safety reasoning and it's explicit attribution to the base model.

Laboratories are prone to severe injuries from minor unsafe actions, yet continuous safety monitoring -- beyond mandatory pre-lab safety training -- is limited by human availability. Vision language models (VLMs) offer promise for autonomous laboratory safety monitoring, but their effectiveness in realistic settings is unclear due to the lack of visual evaluation data, as most safety incidents are documented primarily as unstructured text. To address this gap, we first introduce a structured data generation pipeline that converts textual laboratory scenarios into aligned triples of (image, scene graph, ground truth), using large language models as scene graph architects and image generation models as renderers. Our experiments on the synthetic dataset of 1,207 samples across 362 unique scenarios and seven open- and closed-source models show that VLMs perform effectively given textual scene graph, but degrade substantially in visual-only settings indicating difficulty in extracting structured object relationships directly from pixels. To overcome this, we propose a post-training context-engineering approach, scene-graph-guided alignment, to bridge perceptual gaps in VLMs by translating visual inputs into structured scene graphs better aligned with VLM reasoning, improving hazard detection performance in visual only settings.

As increasingly capable open-weight large language models (LLMs) are deployed, improving their tamper resistance against unsafe modifications, whether accidental or intentional, becomes critical to minimize risks. However, there is no standard approach to evaluate tamper resistance. Varied data sets, metrics, and tampering configurations make it difficult to compare safety, utility, and robustness across different models and defenses. To this end, we introduce TamperBench, the first unified framework to systematically evaluate the tamper resistance of LLMs. TamperBench (i) curates a repository of state-of-the-art weight-space fine-tuning attacks and latent-space representation attacks; (ii) enables realistic adversarial evaluation through systematic hyperparameter sweeps per attack-model pair; and (iii) provides both safety and utility evaluations. TamperBench requires minimal additional code to specify any fine-tuning configuration, alignment-stage defense method, and metric suite while ensuring end-to-end reproducibility. We use TamperBench to evaluate 21 open-weight LLMs, including defense-augmented variants, across nine tampering threats using standardized safety and capability metrics with hyperparameter sweeps per model-attack pair. This yields novel insights, including effects of post-training on tamper resistance, that jailbreak-tuning is typically the most severe attack, and that Triplet emerges as a leading alignment-stage defense. Code is available at: https://github.com/criticalml-uw/TamperBench

Aligning large language models (LLMs) with human values is an increasingly critical step in post-training. Direct Preference Optimization (DPO) has emerged as a simple, yet effective alternative to reinforcement learning from human feedback (RLHF). Synthetic preference data with its low cost and high quality enable effective alignment through single- or multi-model generated preference data. Our study reveals a striking, safety-specific phenomenon associated with DPO alignment: Although multi-model generated data enhances performance on general tasks (ARC, Hellaswag, MMLU, TruthfulQA, Winogrande) by providing diverse responses, it also tends to facilitate reward hacking during training. This can lead to a high attack success rate (ASR) when models encounter jailbreaking prompts. The issue is particularly pronounced when employing stronger models like GPT-4o or larger models in the same family to generate chosen responses paired with target model self-generated rejected responses, resulting in dramatically poorer safety outcomes. Furthermore, with respect to safety, using solely self-generated responses (single-model generation) for both chosen and rejected pairs significantly outperforms configurations that incorporate responses from stronger models, whether used directly as chosen data or as part of a multi-model response pool. We demonstrate that multi-model preference data exhibits high linear separability between chosen and rejected responses, which allows models to exploit superficial cues rather than internalizing robust safety constraints. Our experiments, conducted on models from the Llama, Mistral, and Qwen families, consistently validate these findings.

We introduce SafeWork-R1, a cutting-edge multimodal reasoning model that demonstrates the coevolution of capabilities and safety. It is developed by our proposed SafeLadder framework, which incorporates large-scale, progressive, safety-oriented reinforcement learning post-training, supported by a suite of multi-principled verifiers. Unlike previous alignment methods such as RLHF that simply learn human preferences, SafeLadder enables SafeWork-R1 to develop intrinsic safety reasoning and self-reflection abilities, giving rise to safety `aha' moments. Notably, SafeWork-R1 achieves an average improvement of 46.54% over its base model Qwen2.5-VL-72B on safety-related benchmarks without compromising general capabilities, and delivers state-of-the-art safety performance compared to leading proprietary models such as GPT-4.1 and Claude Opus 4. To further bolster its reliability, we implement two distinct inference-time intervention methods and a deliberative search mechanism, enforcing step-level verification. Finally, we further develop SafeWork-R1-InternVL3-78B, SafeWork-R1-DeepSeek-70B, and SafeWork-R1-Qwen2.5VL-7B. All resulting models demonstrate that safety and capability can co-evolve synergistically, highlighting the generalizability of our framework in building robust, reliable, and trustworthy general-purpose AI.

The acquisition of agentic capabilities has transformed LLMs from "knowledge providers" to "action executors", a trend that while expanding LLMs' capability boundaries, significantly increases their susceptibility to malicious use. Previous work has shown that current LLM-based agents execute numerous malicious tasks even without being attacked, indicating a deficiency in agentic use safety alignment during the post-training phase. To address this gap, we propose AgentAlign, a novel framework that leverages abstract behavior chains as a medium for safety alignment data synthesis. By instantiating these behavior chains in simulated environments with diverse tool instances, our framework enables the generation of highly authentic and executable instructions while capturing complex multi-step dynamics. The framework further ensures model utility by proportionally synthesizing benign instructions through non-malicious interpretations of behavior chains, precisely calibrating the boundary between helpfulness and harmlessness. Evaluation results on AgentHarm demonstrate that fine-tuning three families of open-source models using our method substantially improves their safety (35.8% to 79.5% improvement) while minimally impacting or even positively enhancing their helpfulness, outperforming various prompting methods. The dataset and code have both been open-sourced.

Refusal rates are a poor proxy for LLM safety, i.e., a model may over-refuse benign prompts while still complying with harmful ones. We audit both failure modes across 21 open-weight LLMs on four safety benchmarks (OR-Bench, XSTest, ToxiGen, BOLD), using a composition adjustment to isolate model sensitivity from dataset toxicity confounds. We report three findings. First, models adopt fundamentally different calibration strategies: conservative ecosystems such as Llama suppress unsafe outputs at the cost of elevated over-refusals, while permissive ecosystems such as DeepSeek and Qwen preserve helpfulness but tolerate higher harmful compliance. Second, demographic protection is unequal: models over-protect prominent racial and religious groups, frequently refusing even benign prompts about them, while providing substantially weaker protection against disability-targeted attacks. Third, refusal and compliance tendencies are stable within model families across generations and scales, suggesting that post-training objectives shape safety behavior more than architecture. Our results call for joint, demographically-aware, and multi-judge safety evaluation.

Open-weight AI systems offer unique benefits, including enhanced transparency, open research, and decentralized access. However, they are vulnerable to tampering attacks which can efficiently elicit harmful behaviors by modifying weights or activations. Currently, there is not yet a robust science of open-weight model risk management. Existing safety fine-tuning methods and other post-training techniques have struggled to make LLMs resistant to more than a few dozen steps of adversarial fine-tuning. In this paper, we investigate whether filtering text about dual-use topics from training data can prevent unwanted capabilities and serve as a more tamper-resistant safeguard. We introduce a multi-stage pipeline for scalable data filtering and show that it offers a tractable and effective method for minimizing biothreat proxy knowledge in LLMs. We pretrain multiple 6.9B-parameter models from scratch and find that they exhibit substantial resistance to adversarial fine-tuning attacks on up to 10,000 steps and 300M tokens of biothreat-related text -- outperforming existing post-training baselines by over an order of magnitude -- with no observed degradation to unrelated capabilities. However, while filtered models lack internalized dangerous knowledge, we find that they can still leverage such information when it is provided in context (e.g., via search tool augmentation), demonstrating a need for a defense-in-depth approach. Overall, these findings help to establish pretraining data curation as a promising layer of defense for open-weight AI systems.

Agentic language models operate in a fundamentally different safety regime than chat models: they must plan, call tools, and execute long-horizon actions where a single misstep, such as accessing files or entering credentials, can cause irreversible harm. Existing alignment methods, largely optimized for static generation and task completion, break down in these settings due to sequential decision-making, adversarial tool feedback, and overconfident intermediate reasoning. We introduce MOSAIC, a post-training framework that aligns agents for safe multi-step tool use by making safety decisions explicit and learnable. MOSAIC structures inference as a plan, check, then act or refuse loop, with explicit safety reasoning and refusal as first-class actions. To train without trajectory-level labels, we use preference-based reinforcement learning with pairwise trajectory comparisons, which captures safety distinctions often missed by scalar rewards. We evaluate MOSAIC zero-shot across three model families, Qwen2.5-7B, Qwen3-4B-Thinking, and Phi-4, and across out-of-distribution benchmarks spanning harmful tasks, prompt injection, benign tool use, and cross-domain privacy leakage. MOSAIC reduces harmful behavior by up to 50%, increases harmful-task refusal by over 20% on injection attacks, cuts privacy leakage, and preserves or improves benign task performance, demonstrating robust generalization across models, domains, and agentic settings.

Reward models are used throughout the post-training of language models to capture nuanced signals from preference data and provide a training target for optimization across instruction following, reasoning, safety, and more domains. The community has begun establishing best practices for evaluating reward models, from the development of benchmarks that test capabilities in specific skill areas to others that test agreement with human preferences. At the same time, progress in evaluation has not been mirrored by the effectiveness of reward models in downstream tasks -- simpler direct alignment algorithms are reported to work better in many cases. This paper introduces RewardBench 2, a new multi-skill reward modeling benchmark designed to bring new, challenging data for accuracy-based reward model evaluation -- models score about 20 points on average lower on RewardBench 2 compared to the first RewardBench -- while being highly correlated with downstream performance. Compared to most other benchmarks, RewardBench 2 sources new human prompts instead of existing prompts from downstream evaluations, facilitating more rigorous evaluation practices. In this paper, we describe our benchmark construction process and report how existing models perform on it, while quantifying how performance on the benchmark correlates with downstream use of the models in both inference-time scaling algorithms, like best-of-N sampling, and RLHF training algorithms like proximal policy optimization.

Closed-loop cooperative driving requires planners that generate realistic multimodal multi-agent trajectories while improving safety and traffic efficiency. Existing diffusion planners can model multimodal behaviors from demonstrations, but they often exhibit weak scene consistency and remain poorly aligned with closed-loop objectives; meanwhile, stable online post-training in reactive multi-agent environments remains difficult. We present Multi-ORFT, which couples scene-conditioned diffusion pre-training with stable online reinforcement post-training. In pre-training, the planner uses inter-agent self-attention, cross-attention, and AdaLN-Zero-based scene conditioning to improve scene consistency and road adherence of joint trajectories. In post-training, we formulate a two-level MDP that exposes step-wise reverse-kernel likelihoods for online optimization, and combine dense trajectory-level rewards with variance-gated group-relative policy optimization (VG-GRPO) to stabilize training. On the WOMD closed-loop benchmark, Multi-ORFT reduces collision rate from 2.04% to 1.89% and off-road rate from 1.68% to 1.36%, while increasing average speed from 8.36 to 8.61 m/s relative to the pre-trained planner, and it outperforms strong open-source baselines including SMART-large, SMART-tiny-CLSFT, and VBD on the primary safety and efficiency metrics. These results show that coupling scene-consistent denoising with stable online diffusion-policy optimization improves the reliability of closed-loop cooperative driving.

This technical report presents K-EXAONE 2.0, an open-weight multilingual foundation model developed by LG AI Research as a step in our effort toward global frontier-scale foundation models. Rather than training from scratch, we upcycle K-EXAONE and expand its architecture, yielding a Mixture-of-Experts (MoE) model with 750B total parameters and approximately 37B activated per token---more than three times the capacity of its predecessor. K-EXAONE 2.0 supports context lengths of up to 256K tokens and expands multilingual coverage from six to ten languages. Its training pipeline combines continual pre-training, difficulty-focused mid-training, and post-training to strengthen reasoning, agentic coding, multilingual capability, and safety grounded in Korean sociocultural contexts. Across nine evaluation categories selected to reflect the conditions of practical use, K-EXAONE 2.0 improves over K-EXAONE and remains competitive with open-weight models, showing its largest gains in agentic coding and long-context understanding and its clearest strengths in long-context retrieval and safety. Released under the Apache 2.0 license, K-EXAONE 2.0 enables the wider AI ecosystem to evaluate, deploy, adapt, and build upon it, while marking the beginning---rather than the endpoint---of our challenge toward the global frontier.

Large reasoning models (LRMs) "think" by generating structured chain-of-thought (CoT) before producing a final answer, yet they still lack the ability to reason critically about safety alignment and are easily biased when a flawed premise is injected into their thought process. We propose RECAP (Robust Safety Alignment via Counter-Aligned Prefilling), a principled reinforcement learning (RL) method for post-training that explicitly teaches models to override flawed reasoning trajectories and reroute to safe and helpful responses. RECAP trains on a mixture of synthetically generated counter-aligned CoT prefills and standard prompts, requires no additional training cost or modifications beyond vanilla reinforcement learning from human feedback (RLHF), and substantially improves safety and jailbreak robustness, reduces overrefusal, and preserves core reasoning capability -- all while maintaining inference token budget. Extensive analysis shows that RECAP-trained models engage in self-reflection more frequently and remain robust under adaptive attacks, preserving safety even after repeated attempts to override their reasoning.

Diffusion models have become a popular choice for decision-making tasks in robotics, and more recently, are also being considered for solving autonomous driving tasks. However, their applications and evaluations in autonomous driving remain limited to simulation-based or laboratory settings. The full strength of diffusion models for large-scale, complex real-world settings, such as End-to-End Autonomous Driving (E2E AD), remains underexplored. In this study, we conducted a systematic and large-scale investigation to unleash the potential of the diffusion models as planners for E2E AD, based on a tremendous amount of real-vehicle data and road testing. Through comprehensive and carefully controlled studies, we identify key insights into the diffusion loss space, trajectory representation, and data scaling that significantly impact E2E planning performance. Moreover, we also provide an effective reinforcement learning post-training strategy to further enhance the safety of the learned planner. The resulting diffusion-based learning framework, Hyper Diffusion Planner} (HDP), is deployed on a real-vehicle platform and evaluated across 6 urban driving scenarios and 200 km of real-world testing, achieving a notable 10x performance improvement over the base model. Our work demonstrates that diffusion models, when properly designed and trained, can serve as effective and scalable E2E AD planners for complex, real-world autonomous driving tasks.

This technical report presents Yi-Lightning, our latest flagship large language model (LLM). It achieves exceptional performance, ranking 6th overall on Chatbot Arena, with particularly strong results (2nd to 4th place) in specialized categories including Chinese, Math, Coding, and Hard Prompts. Yi-Lightning leverages an enhanced Mixture-of-Experts (MoE) architecture, featuring advanced expert segmentation and routing mechanisms coupled with optimized KV-caching techniques. Our development process encompasses comprehensive pre-training, supervised fine-tuning (SFT), and reinforcement learning from human feedback (RLHF), where we devise deliberate strategies for multi-stage training, synthetic data construction, and reward modeling. Furthermore, we implement RAISE (Responsible AI Safety Engine), a four-component framework to address safety issues across pre-training, post-training, and serving phases. Empowered by our scalable super-computing infrastructure, all these innovations substantially reduce training, deployment and inference costs while maintaining high-performance standards. With further evaluations on public academic benchmarks, Yi-Lightning demonstrates competitive performance against top-tier LLMs, while we observe a notable disparity between traditional, static benchmark results and real-world, dynamic human preferences. This observation prompts a critical reassessment of conventional benchmarks' utility in guiding the development of more intelligent and powerful AI systems for practical applications. Yi-Lightning is now available through our developer platform at https://platform.lingyiwanwu.com.

We introduce GPT-Red, an automated red-teaming agent that is trained to discover novel prompt injection attacks against frontier LLMs. The goal of this model is to evaluate and improve the robustness of our production systems. To this end, we use it to adversarially train GPT-5.6, our most robust model to prompt injections to date. To create GPT-Red, we design a scalable self-play algorithm where the model is tasked with attacking a diverse population of simultaneously-trained defender agents. We train the model on realistic red-teaming environments using compute on the same scale as some of our largest RL post-training runs, making it the single-largest LLM safety training run ever documented. GPT-Red excels at red-teaming: it reliably breaks our past models up to GPT-5.5, it finds more successful attacks than human red-teamers, and it generalizes to held-out environments, defender models, and harnesses. In the future, we expect that as we improve the robustness of each new GPT model, it will in turn will provide better learning signal for even stronger red-teamer agents, thus unlocking a self-improvement flywheel.

As LLM agents increasingly select tools autonomously, their choices among tools with different privileges become safety-relevant. However, prior tool-selection studies focus on safety-agnostic metadata preferences, leaving privilege-sensitive choices underexplored. To address this gap, we study over-privileged tool selection, in which an agent selects or escalates to a higher-privilege tool despite a sufficient lower-privilege alternative. We introduce ToolPrivBench to evaluate whether agents choose higher-privilege tools despite sufficient lower-privilege alternatives, measuring both initial selection and escalation after transient tool failures. Across eight domains and five recurring risk patterns, we find that over-privileged tool selection is common among mainstream LLM agents and is further amplified by transient failures. We further find that general safety alignment does not reliably transfer to least-privilege tool choice, while prompt-level controls provide only limited mitigation under transient failures. We therefore introduce a privilege-aware post-training defense that teaches agents to prefer sufficient lower-privilege tools and escalate only when necessary. Our mitigation experiments show that this defense substantially reduces unnecessary high-privilege tool use while preserving general capabilities.

The release of ChatGPT in November 2022 sparked an explosion of interest in post-training and an avalanche of new preference optimization (PO) methods. These methods claim superior alignment by virtue of better correspondence with human pairwise preferences, often measured by LLM judges. In this work, we attempt to answer the following question -- do LLM-judge preferences translate to progress on other, more concrete metrics for alignment, and if not, why not? We define a concrete metric for alignment, and introduce SOS-Bench, the largest standardized, reproducible LLM meta-benchmark to date. We find that (1) LLM-judgments do not correlate with concrete measures of safety, world knowledge, and instruction following; (2) LLM judges have powerful implicit biases, prioritizing style over factuality and safety; and (3) the supervised fine-tuning (SFT) stage of post-training, and not the PO stage, has the greatest impact on alignment, with data scaling and prompt diversity as the driving factors. Our codebase and complete results can be found at https://github.com/penfever/sos-bench.

Large language models have achieved remarkable capabilities across diverse tasks, yet their internal decision-making processes remain largely opaque, limiting our ability to inspect, control, and systematically improve them. This opacity motivates a growing body of research in mechanistic interpretability, with sparse autoencoders (SAEs) emerging as one of the most promising tools for decomposing model activations into sparse, interpretable feature representations. We introduce Qwen-Scope, an open-source suite of SAEs built on the Qwen model family, comprising 14 groups of SAEs across 7 model variants from the Qwen3 and Qwen3.5 series, covering both dense and mixture-of-expert architectures. Built on top of these SAEs, we show that SAEs can go beyond post-hoc analysis to serve as practical interfaces for model development along four directions: (i) inference-time steering, where SAE feature directions control language, concepts, and preferences without modifying model weights; (ii) evaluation analysis, where activated SAE features provide a representation-level proxy for benchmark redundancy and capability coverage; (iii) data-centric workflows, where SAE features support multilingual toxicity classification and safety-oriented data synthesis; and (iv) post-training optimization, where SAE-derived signals are incorporated into supervised fine-tuning and reinforcement learning objectives to mitigate undesirable behaviors such as code-switching and repetition. Together, these results demonstrate that SAEs can serve not only as post-hoc analysis tools, but also as reusable representation-level interfaces for diagnosing, controlling, evaluating, and improving large language models. By open-sourcing Qwen-Scope, we aim to support mechanistic research and accelerate practical workflows that connect model internals to downstream behavior.

Ensuring safety alignment is a critical requirement for large language models (LLMs), particularly given increasing deployment in real-world applications. Despite considerable advancements, LLMs remain susceptible to jailbreak attacks, which exploit system vulnerabilities to circumvent safety measures and elicit harmful or inappropriate outputs. Furthermore, while adversarial training-based defense methods have shown promise, a prevalent issue is the unintended over-defense behavior, wherein models excessively reject benign queries, significantly undermining their practical utility. To address these limitations, we introduce LATPC, a Latent-space Adversarial Training with Post-aware Calibration framework. LATPC dynamically identifies safety-critical latent dimensions by contrasting harmful and benign inputs, enabling the adaptive construction of targeted refusal feature removal attacks. This mechanism allows adversarial training to concentrate on real-world jailbreak tactics that disguise harmful queries as benign ones. During inference, LATPC employs an efficient embedding-level calibration mechanism to minimize over-defense behaviors with negligible computational overhead. Experimental results across five types of disguise-based jailbreak attacks demonstrate that LATPC achieves a superior balance between safety and utility compared to existing defense frameworks. Further analysis demonstrates the effectiveness of leveraging safety-critical dimensions in developing robust defense methods against jailbreak attacks.

Fine-tuning on harmless data can partially undo behaviors acquired earlier in training. Safety can erode under benign post-alignment updates, unlearned capabilities can re-emerge, latent traits can transfer through apparently unrelated supervision, and related post-alignment fragility appears in other generative settings. We argue these phenomena are usefully viewed through a common training-history lens. Our hypothesis is geometric: large early training phases create dominant behavioral manifolds, while later alignment or specialization phases are shallower displacements from them. Subsequent fine-tuning can therefore inherit a persistent reversion component pointing back toward a witness of the dominant manifold. We call this the gravitational interpretation of fine-tuning reversion. Across our main settings, representational drift rapidly acquires a component along a history-defined reversion direction (v_rev). In our main track, alignment with v_rev rises from cos = 0.429 +/- 0.052 after the first update to 0.647 +/- 0.021 by step 20. Across 24 run-step pairs, every observed alignment exceeds the p99 of an isotropic activation-space null. We demonstrate that selectively blocking motion along v_rev changes the final alignment at T=100 from 0.648 +/- 0.009 to -0.211 +/- 0.021 and reduces harmfulness from 19.0% +/- 4.0% to 8.5% +/- 1.5% with little task cost. These results support v_rev as a causally relevant mediator of early post-alignment reversion in our setup. Importantly, we do not claim that v_rev is the unique safety direction, nor that the dominant manifold is directly observed; rather, we identify a robust, history-defined direction that explains and partially controls early reversion dynamics.

Safety aligned Large Language Models (LLMs) are vulnerable to harmful fine-tuning attacks qi2023fine-- a few harmful data mixed in the fine-tuning dataset can break the LLMs's safety alignment. Existing mitigation strategies include alignment stage solutions huang2024vaccine, rosati2024representation and fine-tuning stage solutions huang2024lazy,mukhoti2023fine. However, our evaluation shows that both categories of defenses fail when some specific training hyper-parameters are chosen -- a large learning rate or a large number of training epochs in the fine-tuning stage can easily invalidate the defense, which however, is necessary to guarantee finetune performance. To this end, we propose Antidote, a post-fine-tuning stage solution, which remains \textit{agnostic to the training hyper-parameters in the fine-tuning stage}. Antidote relies on the philosophy that by removing the harmful parameters, the harmful model can be recovered from the harmful behaviors, regardless of how those harmful parameters are formed in the fine-tuning stage. With this philosophy, we introduce a one-shot pruning stage after harmful fine-tuning to remove the harmful weights that are responsible for the generation of harmful content. Despite its embarrassing simplicity, empirical results show that Antidote can reduce harmful score while maintaining accuracy on downstream tasks.Our project page is at https://huangtiansheng.github.io/Antidote_gh_page/

Current methods for content safety in Large Language Models (LLMs), such as Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF), often rely on multi-stage training pipelines and lack fine-grained, post-deployment controllability. To address these limitations, we propose a unified co-training framework that efficiently integrates multiple safety behaviors: positive (lawful/prosocial), negative (unfiltered/risk-prone) and rejective (refusal-oriented/conservative) within a single SFT stage. Notably, each behavior is dynamically activated via a simple system-level instruction, or magic token, enabling stealthy and efficient behavioral switching at inference time. This flexibility supports diverse deployment scenarios, such as positive for safe user interaction, negative for internal red-teaming, and rejective for context-aware refusals triggered by upstream moderation signals. This co-training strategy induces a distinct Safety Alignment Margin in the output space, characterized by well-separated response distributions corresponding to each safety mode. The existence of this margin provides empirical evidence for the model's safety robustness and enables unprecedented fine-grained control. Experiments show that our method matches the safety alignment quality of SFT+DPO, with our 8B model notably surpassing DeepSeek-R1 (671B) in safety performance, while significantly reducing both training complexity and deployment costs. This work presents a scalable, efficient, and highly controllable solution for LLM content safety.

What would it mean to have a scientific understanding of AI? Models are not static objects: they are snapshots of time-evolving processes shaped by data, objectives, architectures, and optimization dynamics. Yet much of AI research treats models as fixed artifacts, analyzing behaviors after training rather than asking why they emerge. This position paper argues that a science of AI must move beyond post-hoc fixes and study the training dynamics that produce model behavior. Such a science should support progressively stronger forms of understanding: predicting outcomes from early training signals, intervening when trajectories go wrong, and ultimately designing training procedures that more reliably produce desired properties. Scaling laws have made prediction routine for loss; the challenge is extending this success to capabilities, biases, robustness, and safety-relevant behaviors. We articulate requirements for such theories grounded in the history and philosophy of science, examine progress in mechanistic interpretability, fairness, memorization, and simplicity bias, and identify concrete open problems.

In this work, we show that it is possible to extract significant amounts of alignment training data from a post-trained model -- useful to steer the model to improve certain capabilities such as long-context reasoning, safety, instruction following, and maths. While the majority of related work on memorisation has focused on measuring success of training data extraction through string matching, we argue that embedding models are better suited for our specific goals. Distances measured through a high quality embedding model can identify semantic similarities between strings that a different metric such as edit distance will struggle to capture. In fact, in our investigation, approximate string matching would have severely undercounted (by a conservative estimate of 10times) the amount of data that can be extracted due to trivial artifacts that deflate the metric. Interestingly, we find that models readily regurgitate training data that was used in post-training phases such as SFT or RL. We show that this data can be then used to train a base model, recovering a meaningful amount of the original performance. We believe our work exposes a possibly overlooked risk towards extracting alignment data. Finally, our work opens up an interesting discussion on the downstream effects of distillation practices: since models seem to be regurgitating aspects of their training set, distillation can therefore be thought of as indirectly training on the model's original dataset.

Chain-of-Thought (CoT) monitoring, in which automated systems monitor the CoT of an LLM, is a promising approach for effectively overseeing AI systems. However, the extent to which a model's CoT helps us oversee the model - the monitorability of the CoT - can be affected by training, for instance by the model learning to hide important features of its reasoning. We propose and empirically validate a conceptual framework for predicting when and why this occurs. We model LLM post-training as an RL environment where the reward decomposes into two terms: one term depending on final outputs and another term depending on the CoT. Our framework allows us to classify these two terms as "aligned", "orthogonal", or "in-conflict" before training. We predict that training with in-conflict terms will reduce monitorability, orthogonal terms will not affect it, and aligned terms will improve it. To validate our framework, we use it to classify a set of RL environments, train LLMs within those environments, and evaluate how training affects CoT monitorability. We find that (1) training with "in-conflict" reward terms reduces CoT monitorability and (2) optimizing in-conflict reward terms is difficult.

Recent advancements in Large Language Models (LLMs) demonstrate remarkable capabilities across various fields. These developments have led to more direct communication between humans and LLMs in various situations, such as social companionship and psychological support. However, LLMs often exhibit limitations in emotional perception and social competence during real-world conversations. These limitations partly originate from their inability to adapt their communication style and emotional expression to different social and task contexts. In this work, we introduce PersonaFuse, a novel LLM post-training framework that enables LLMs to adapt and express different personalities for varying situations. Inspired by Trait Activation Theory and the Big Five personality model, PersonaFuse employs a Mixture-of-Expert architecture that combines persona adapters with a dynamic routing network, enabling contextual trait expression. Experimental results show that PersonaFuse substantially outperforms baseline models across multiple dimensions of social-emotional intelligence. Importantly, these gains are achieved without sacrificing general reasoning ability or model safety, which remain common limitations of direct prompting and supervised fine-tuning approaches. PersonaFuse also delivers consistent improvements in downstream human-centered applications, such as mental health counseling and review-based customer service. Finally, human preference evaluations against leading LLMs, including GPT-4o and DeepSeek, demonstrate that PersonaFuse achieves competitive response quality despite its comparatively smaller model size. These findings demonstrate that PersonaFuse~offers a theoretically grounded and practical approach for developing social-emotional enhanced LLMs, marking a significant advancement toward more human-centric AI systems.

Reinforcement learning (RL) has become a dominant post-training paradigm, enabling large language models (LLMs) to learn from rewards. We observe that societal regulations are structurally similar to reward functions. They define measurable outcomes, thresholds, and exceptions, while often leaving institutional intent only partially specified. We hypothesise that the RL training process may exploit these gaps and therefore ask whether models' well-known tendency to hack reward functions during RL can scale into a more consequential failure mode named societal hacking: discovering loopholes in the rules society runs on. To study this phenomenon, we introduce SocioHack, a sandbox of 72 societal environments, and find that within these environments, reward hacking naturally emerges and leads to regulatory loophole discovery. Models learn to hack the social rules and generate strategies that remain technically compliant while defeating regulatory intent, and current LLM safeguards provide only limited mitigation. Therefore, collecting in-the-wild feedback for model training requires greater caution, and we need a next-generation post-training paradigm for safely iterating LLMs in real society.=

Safety is critical to broadening the application of reinforcement learning (RL). Often, we train RL agents in a controlled environment, such as a laboratory, before deploying them in the real world. However, the real-world target task might be unknown prior to deployment. Reward-free RL trains an agent without the reward to adapt quickly once the reward is revealed. We consider the constrained reward-free setting, where an agent (the guide) learns to explore safely without the reward signal. This agent is trained in a controlled environment, which allows unsafe interactions and still provides the safety signal. After the target task is revealed, safety violations are not allowed anymore. Thus, the guide is leveraged to compose a safe behaviour policy. Drawing from transfer learning, we also regularize a target policy (the student) towards the guide while the student is unreliable and gradually eliminate the influence of the guide as training progresses. The empirical analysis shows that this method can achieve safe transfer learning and helps the student solve the target task faster.

As large language models (LLMs) are increasingly deployed in high-stakes settings, the risk of generating harmful or toxic content remains a central challenge. Post-hoc alignment methods are brittle: once unsafe patterns are learned during pretraining, they are hard to remove. We present a data-centric pretraining framework that builds safety into the model from the start. Our contributions include: (i) a safety classifier trained on 10,000 GPT-4 labeled examples, used to filter 600B tokens; (ii) the largest synthetic safety dataset to date (100B tokens) generated via recontextualization of harmful web data; (iii) RefuseWeb and Moral Education datasets that convert harmful prompts into refusal dialogues and web-style educational material; (iv) Harmfulness-Tag annotations injected during pretraining to flag unsafe content and steer away inference from harmful generations; and (v) safety evaluations measuring base model behavior before instruction tuning. Our safety-pretrained models reduce attack success rates from 38.8% to 8.4% with no performance degradation on standard LLM safety benchmarks.

State-of-the-art AI systems can be significantly improved without expensive retraining via "post-training enhancements"-techniques applied after initial training like fine-tuning the system to use a web browser. We review recent post-training enhancements, categorizing them into five types: tool-use, prompting methods, scaffolding, solution selection, and data generation. Different enhancements improve performance on different tasks, making it hard to compare their significance. So we translate improvements from different enhancements into a common currency, the compute-equivalent gain: how much additional training compute would be needed to improve performance by the same amount as the enhancement. Our non-experimental work shows that post-training enhancements have significant benefits: most surveyed enhancements improve benchmark performance by more than a 5x increase in training compute, some by more than 20x. Post-training enhancements are relatively cheap to develop: fine-tuning costs are typically <1% of the original training cost. Governing the development of capable post-training enhancements may be challenging because frontier models could be enhanced by a wide range of actors.

Post-training, which elicits a pretrained Base model into the corresponding Instruct model, is widely considered to be superficial. In this work, we first reinforce this hypothesis by providing novel quantitative evidence from the weight level that the effective rank (eRank) remains negligibly changed. However, this superficiality also suffers a critical trade-off, improving the exploitation capabilities at the cost of limiting its exploration. To tackle this issue, we propose Timber, a simple yet effective training-free method that enhances the exploration capability of the Instruct model while preserving its exploitation. The key insight is to partially revert Instruct towards the paired Base model by subtle yet targeted refinement of the weight deltas. Extensive experiments on Llama and Qwen series demonstrate that Timber consistently improves vanilla Instruct models, particularly on Pass@k performance. Our findings offer new insights into the post-training stage at the weight level and practical strategies to refine the Instruct model without training.

Laboratory accidents pose significant risks to human life and property, underscoring the importance of robust safety protocols. Despite advancements in safety training, laboratory personnel may still unknowingly engage in unsafe practices. With the increasing reliance on large language models (LLMs) for guidance in various fields, including laboratory settings, there is a growing concern about their reliability in critical safety-related decision-making. Unlike trained human researchers, LLMs lack formal lab safety education, raising questions about their ability to provide safe and accurate guidance. Existing research on LLM trustworthiness primarily focuses on issues such as ethical compliance, truthfulness, and fairness but fails to fully cover safety-critical real-world applications, like lab safety. To address this gap, we propose the Laboratory Safety Benchmark (LabSafety Bench), a comprehensive evaluation framework based on a new taxonomy aligned with Occupational Safety and Health Administration (OSHA) protocols. This benchmark includes 765 multiple-choice questions verified by human experts, assessing LLMs and vision language models (VLMs) performance in lab safety contexts. Our evaluations demonstrate that while GPT-4o outperforms human participants, it is still prone to critical errors, highlighting the risks of relying on LLMs in safety-critical environments. Our findings emphasize the need for specialized benchmarks to accurately assess the trustworthiness of LLMs in real-world safety applications.

Finetuning large language models (LLMs) enables user-specific customization but introduces critical safety risks: even a few harmful examples can compromise safety alignment. A common mitigation strategy is to update the model more strongly on examples deemed safe, while downweighting or excluding those flagged as unsafe. However, because safety context can shift within a single example, updating the model equally on both harmful and harmless parts of a response is suboptimal-a coarse treatment we term static safety shaping. In contrast, we propose dynamic safety shaping (DSS), a framework that uses fine-grained safety signals to reinforce learning from safe segments of a response while suppressing unsafe content. To enable such fine-grained control during finetuning, we introduce a key insight: guardrail models, traditionally used for filtering, can be repurposed to evaluate partial responses, tracking how safety risk evolves throughout the response, segment by segment. This leads to the Safety Trajectory Assessment of Response (STAR), a token-level signal that enables shaping to operate dynamically over the training sequence. Building on this, we present STAR-DSS, guided by STAR scores, that robustly mitigates finetuning risks and delivers substantial safety improvements across diverse threats, datasets, and model families-all without compromising capability on intended tasks. We encourage future safety research to build on dynamic shaping principles for stronger mitigation against evolving finetuning risks.

Workplace accidents continue to pose significant risks for human safety, particularly in industries such as construction and manufacturing, and the necessity for effective Personal Protective Equipment (PPE) compliance has become increasingly paramount. Our research focuses on the development of non-invasive techniques based on the Object Detection (OD) and Convolutional Neural Network (CNN) to detect and verify the proper use of various types of PPE such as helmets, safety glasses, masks, and protective clothing. This study proposes the SH17 Dataset, consisting of 8,099 annotated images containing 75,994 instances of 17 classes collected from diverse industrial environments, to train and validate the OD models. We have trained state-of-the-art OD models for benchmarking, and initial results demonstrate promising accuracy levels with You Only Look Once (YOLO)v9-e model variant exceeding 70.9% in PPE detection. The performance of the model validation on cross-domain datasets suggests that integrating these technologies can significantly improve safety management systems, providing a scalable and efficient solution for industries striving to meet human safety regulations and protect their workforce. The dataset is available at https://github.com/ahmadmughees/sh17dataset.

Safe Reinforcement Learning (RL) algorithms are typically evaluated under fixed training conditions. We investigate whether training-time safety guarantees transfer to deployment under distribution shift, using diabetes management as a safety-critical testbed. We benchmark safe RL algorithms on a unified clinical simulator and reveal a safety generalization gap: policies satisfying constraints during training frequently violate safety requirements on unseen patients. We demonstrate that test-time shielding, which filters unsafe actions using learned dynamics models, effectively restores safety across algorithms and patient populations. Across eight safe RL algorithms, three diabetes types, and three age groups, shielding achieves Time-in-Range gains of 13--14\% for strong baselines such as PPO-Lag and CPO while reducing clinical risk index and glucose variability. Our simulator and benchmark provide a platform for studying safety under distribution shift in safety-critical control domains. Code is available at https://github.com/safe-autonomy-lab/GlucoSim and https://github.com/safe-autonomy-lab/GlucoAlg.

Text-to-image models trained on large-scale data often inevitably ingest unsafe content. While some people observe input-output amplifications, it remains unclear whether and how training data composition directly drives model output safety or by other factors. We shed light on this question by isolating this variable: we train the same text-to-image model on datasets that differ only in their fraction of unsafe images (0\% to 9.6\%), across several dataset scales (100K to 8M). Then we generate images with the resulting models, and evaluate them with four independent safety classifiers. Output unsafety rises monotonically from 16.6\% at 0\% contamination to 25.5\% at 5\%. A factorial design reveals that the proportion, not the absolute count, of unsafe training images is the operative variable. The 16.6\% irreducible baseline at zero contamination implicates the other components, e.g. frozen text encoder, as a residual safety risk -- confirmed by a text encoder ablation showing that SafeCLIP reduces this floor to 9.6\%, while the dose-response effect persists across all three encoders tested. Critically, no quality degradation in terms of FID, CLIPscore and ImageReward accompanies safety filtering. These results establish that data curation and text encoder safety are complementary and independently effective interventions. At the same time, the remaining level of unsafety poses questions for future research about emerging capabilities and compositionality.

Post-training has become a primary driver of recent progress in large reasoning models, and reasoning data are often the key variable determining whether this stage succeeds. Work on post-training reasoning data has grown rapidly, yet this literature remains scattered across dataset papers, reinforcement-learning recipes, reward-model studies, benchmarks, and frontier system reports. This paper is the first primer to synthesize over 150 key public studies and system reports on post-training reasoning data. We organize the field around four questions: what data objects exist, what makes them useful, how they are constructed, and how they scale. Together, this organization provides an attribution framework for future reasoning-data releases and post-training recipes.

We present an empirical benchmark evaluating how five large language models assess multisensor physical hazard data. Testing 60 scenarios across three categories - multi-sensor joint assessment, response proportionality, and pattern disambiguation - with 1,800 API calls at temperature 0.0, we find that all tested models consistently produced no precautionary warning signal across the tested scenarios where multiple sensors are simultaneously elevated below their individual safety limits, while achieving near-perfect accuracy on single-sensor threshold violations. All five models (ChatGPT-4o, Gemini 2.5 Flash, DeepSeek, Kimi, Llama 3.1 8B) score near zero on Category A multi-sensor scenarios (Q2: 0.000-0.208; Q3: 0.000-0.592) compared to strong performance on single-sensor scenarios (Category B Q1: 0.975-1.000). Structured tabular formatting shows no consistent advantage over plain prose; ChatGPT-4o performs significantly better under prose (p = 0.001). These findings have direct implications for practitioners deploying the tested models in physical safety monitoring systems.

AI agents have become surprisingly proficient at software engineering over the past year, largely due to improvements in reasoning capabilities. This raises a deeper question: can these systems extend their capabilities to automate AI research itself? In this paper, we explore post-training, the critical phase that turns base LLMs into useful assistants. We introduce PostTrainBench to benchmark how well LLM agents can perform post-training autonomously under bounded compute constraints (10 hours on one H100 GPU). We ask frontier agents (e.g., Claude Code with Opus 4.6) to optimize the performance of a base LLM on a particular benchmark (e.g., Qwen3-4B on AIME). Importantly, we do not provide any predefined strategies to the agents and instead give them full autonomy to find necessary information on the web, run experiments, and curate data. We find that frontier agents make substantial progress but generally lag behind instruction-tuned LLMs from leading providers: 23.2% for the best agent vs. 51.1% for official instruction-tuned models. However, agents can exceed instruction-tuned models in targeted scenarios: GPT-5.1 Codex Max achieves 89% on BFCL with Gemma-3-4B vs. 67% for the official model. We also observe several failure modes worth flagging. Agents sometimes engage in reward hacking: training on the test set, downloading existing instruction-tuned checkpoints instead of training their own, and using API keys they find to generate synthetic data without authorization. These behaviors are concerning and highlight the importance of careful sandboxing as these systems become more capable. Overall, we hope PostTrainBench will be useful for tracking progress in AI R&D automation and for studying the risks that come with it. Website and code are available at https://posttrainbench.com/.

Safety and responsibility evaluations of advanced AI models are a critical but developing field of research and practice. In the development of Google DeepMind's advanced AI models, we innovated on and applied a broad set of approaches to safety evaluation. In this report, we summarise and share elements of our evolving approach as well as lessons learned for a broad audience. Key lessons learned include: First, theoretical underpinnings and frameworks are invaluable to organise the breadth of risk domains, modalities, forms, metrics, and goals. Second, theory and practice of safety evaluation development each benefit from collaboration to clarify goals, methods and challenges, and facilitate the transfer of insights between different stakeholders and disciplines. Third, similar key methods, lessons, and institutions apply across the range of concerns in responsibility and safety - including established and emerging harms. For this reason it is important that a wide range of actors working on safety evaluation and safety research communities work together to develop, refine and implement novel evaluation approaches and best practices, rather than operating in silos. The report concludes with outlining the clear need to rapidly advance the science of evaluations, to integrate new evaluations into the development and governance of AI, to establish scientifically-grounded norms and standards, and to promote a robust evaluation ecosystem.

Large Language Models used in ChatGPT have traditionally been trained to learn a refusal boundary: depending on the user's intent, the model is taught to either fully comply or outright refuse. While this is a strong mitigation for explicitly malicious prompts, focusing safety training on refusals can lead to brittleness for prompts with obscured user intent. Binary refusal boundaries are especially ill-suited for dual-use cases (such as biology or cybersecurity), where a user request can be answered safely at a high level, but in some cases can lead to malicious uplift if sufficiently detailed or actionable. As an alternative, we propose safe-completions: a safety-training approach that centers on the safety of the assistant's output, rather than a binary classification of the user's intent. Safe-completions seek to maximize helpfulness within the safety policy's constraints. We incorporated this approach into GPT-5 and find that across both production comparisons and internally controlled experiments, safe-completion training improves safety (especially on dual-use prompts), reduces the severity of residual safety failures, and substantially increases model helpfulness.

Multimodal safety moderation requires distinguishing risks arising from visual content, user intent, and assistant behavior. Existing safeguards, however, are typically trained for a single judgment target and reduce safety assessment to a binary decision. Consequently, risk becomes difficult to compare across a multimodal interaction, and ambiguous cases are obscured. We introduce SafeAtlas-VL, a dataset of 1.5M training instances that places image-, request-, and response-level judgments on a five-level ordered scale. We curate a broad collection of safety-relevant data from both real-world and synthetic sources and apply a disagreement-aware annotation procedure. The resulting dataset spans 15 harm categories and 55 fine-grained subcategories, covering a broad range of multimodal safety scenarios. We also construct SafeAtlas-Bench, a held-out set of 5,000 instances for evaluating five-level predictions and continuous risk scores. Upon this dataset, we train the SafeAtlas Guard series of models via target-conditioned tuning for multimodal safety detection. Our models not only perform five-way classification of safety levels but also map safety to continuous scores through a soft cumulative ordinal head. Experimental results demonstrate that guard models trained on our dataset exhibit strong generalization: even without using the training sets of other benchmarks, they achieve competitive performance on the corresponding test sets. Notably, our 8B model attains the overall best performance, outperforming the previous SOTA by approximately 4% in F1 score. Code, data, and models are released to support further research. Warning: this paper contains example data that may be offensive, harmful, graphic, or disturbing.

Effective training and debriefing are critical in high-stakes, mission-critical environments such as disaster response, military simulations, and industrial safety, where precision and minimizing errors are paramount. The traditional post-training analysis relies on manually reviewing 2D videos, a time-consuming process that lacks comprehensive situational awareness. To address these limitations, we introduce ACT360, a system that leverages 360-degree videos and machine learning for automated action detection and structured debriefing. ACT360 integrates 360YOWO, an enhanced You Only Watch Once (YOWO) model with spatial attention and equirectangular-aware convolution (EAC) to mitigate panoramic video distortions. To enable deployment in resource-constrained environments, we apply quantization and model pruning, reducing the model size by 74% while maintaining robust accuracy (mAP drop of only 1.5%, from 0.865 to 0.850) and improving inference speed. We validate our approach on a publicly available dataset of 55 labeled 360-degree videos covering seven key operational actions, recorded across various real-world training sessions and environmental conditions. Additionally, ACT360 integrates 360AIE (Action Insight Explorer), a web-based interface for automatic action detection, retrieval, and textual summarization using large language models (LLMs), significantly enhancing post-incident analysis efficiency. ACT360 serves as a generalized framework for mission-critical debriefing, incorporating EAC, spatial attention, summarization, and model optimization. These innovations apply to any training environment requiring lightweight action detection and structured post-exercise analysis.

Post-training is essential for the success of large language models (LLMs), transforming pre-trained base models into more useful and aligned post-trained models. While plenty of works have studied post-training algorithms and evaluated post-training models by their outputs, it remains understudied how post-training reshapes LLMs internally. In this paper, we compare base and post-trained LLMs mechanistically from four perspectives to better understand post-training effects. Our findings across model families and datasets reveal that: (1) Post-training does not change the factual knowledge storage locations, and it adapts knowledge representations from the base model while developing new knowledge representations; (2) Both truthfulness and refusal can be represented by linear vectors in the hidden representation space. The truthfulness direction is highly similar between the base and post-trained model, and it is effectively transferable for interventions; (3) The refusal direction is different between the base and post-trained models, and it shows limited forward transferability; (4) Differences in confidence between the base and post-trained models cannot be attributed to entropy neurons. Our study provides insights into the fundamental mechanisms preserved and altered during post-training, facilitates downstream tasks like model steering, and could potentially benefit future research in interpretability and LLM post-training.

As Large Language Models (LLMs) and generative AI become increasingly widespread, concerns about content safety have grown in parallel. Currently, there is a clear lack of high-quality, human-annotated datasets that address the full spectrum of LLM-related safety risks and are usable for commercial applications. To bridge this gap, we propose a comprehensive and adaptable taxonomy for categorizing safety risks, structured into 12 top-level hazard categories with an extension to 9 fine-grained subcategories. This taxonomy is designed to meet the diverse requirements of downstream users, offering more granular and flexible tools for managing various risk types. Using a hybrid data generation pipeline that combines human annotations with a multi-LLM "jury" system to assess the safety of responses, we obtain Aegis 2.0, a carefully curated collection of 34,248 samples of human-LLM interactions, annotated according to our proposed taxonomy. To validate its effectiveness, we demonstrate that several lightweight models, trained using parameter-efficient techniques on Aegis 2.0, achieve performance competitive with leading safety models fully fine-tuned on much larger, non-commercial datasets. In addition, we introduce a novel training blend that combines safety with topic following data.This approach enhances the adaptability of guard models, enabling them to generalize to new risk categories defined during inference. We plan to open-source Aegis 2.0 data and models to the research community to aid in the safety guardrailing of LLMs.

Recent progress in generative modeling has made safety control a central challenge, yet existing approaches remain largely model-specific, requiring retraining or tailored interventions for each new architecture. In this work, we ask whether safety can be represented as a portable latent direction, learned once and reused across heterogeneous generators. We introduce the first framework for cross-model safety steering, in which a safety direction is estimated in a source LLM from paired safe-unsafe prompts, transported to a target generator through a lightweight alignment fitted on benign data alone, and applied at inference time. Crucially, our pipeline never accesses unsafe data on the target side, isolating whether safety can be transferred through shared representation geometry. Beyond a single global direction, we also identify a multi-vector extension that captures category-specific safety behaviors, enabling more selective control. We evaluate our approach in text-to-image and text-to-video generation across diverse source-target model pairs. Across models, transferred safety directions achieve ASR reduction and CLIP-Score/FID trade-offs comparable to directions learned natively on the target model using unsafe data, while requiring no target-side unsafe data. This indicates that safety improvements do not come at the expense of generation quality. Our results point to a modular view of safety: safety-relevant behavior is not purely model-local, but can be controlled through latent directions that persist across models. This suggests a new path toward lightweight, reusable safety mechanisms that do not require target-side unsafe data.

Agents should avoid unsafe behaviour during both training and deployment. This typically requires a simulator and a procedural specification of unsafe behaviour. Unfortunately, a simulator is not always available, and procedurally specifying constraints can be difficult or impossible for many real-world tasks. A recently introduced technique, ReQueST, aims to solve this problem by learning a neural simulator of the environment from safe human trajectories, then using the learned simulator to efficiently learn a reward model from human feedback. However, it is yet unknown whether this approach is feasible in complex 3D environments with feedback obtained from real humans - whether sufficient pixel-based neural simulator quality can be achieved, and whether the human data requirements are viable in terms of both quantity and quality. In this paper we answer this question in the affirmative, using ReQueST to train an agent to perform a 3D first-person object collection task using data entirely from human contractors. We show that the resulting agent exhibits an order of magnitude reduction in unsafe behaviour compared to standard reinforcement learning.

Training large language models to follow instructions makes them perform better on a wide range of tasks, generally becoming more helpful. However, a perfectly helpful model will follow even the most malicious instructions and readily generate harmful content. In this paper, we raise concerns over the safety of models that only emphasize helpfulness, not safety, in their instruction-tuning. We show that several popular instruction-tuned models are highly unsafe. Moreover, we show that adding just 3% safety examples (a few hundred demonstrations) in the training set when fine-tuning a model like LLaMA can substantially improve their safety. Our safety-tuning does not make models significantly less capable or helpful as measured by standard benchmarks. However, we do find a behavior of exaggerated safety, where too much safety-tuning makes models refuse to respond to reasonable prompts that superficially resemble unsafe ones. Our study sheds light on trade-offs in training LLMs to follow instructions and exhibit safe behavior.

Vision-Language adaptation (VL adaptation) transforms Large Language Models (LLMs) into Large Vision-Language Models (LVLMs) for multimodal tasks, but this process often compromises the inherent safety capabilities embedded in the original LLMs. Despite potential harmfulness due to weakened safety measures, in-depth analysis on the effects of VL adaptation on safety remains under-explored. This study examines how VL adaptation influences safety and evaluates the impact of safety fine-tuning methods. Our analysis reveals that safety degradation occurs during VL adaptation, even when the training data is safe. While safety tuning techniques like supervised fine-tuning with safety datasets or reinforcement learning from human feedback mitigate some risks, they still lead to safety degradation and a reduction in helpfulness due to over-rejection issues. Further analysis of internal model weights suggests that VL adaptation may impact certain safety-related layers, potentially lowering overall safety levels. Additionally, our findings demonstrate that the objectives of VL adaptation and safety tuning are divergent, which often results in their simultaneous application being suboptimal. To address this, we suggest the weight merging approach as an optimal solution effectively reducing safety degradation while maintaining helpfulness. These insights help guide the development of more reliable and secure LVLMs for real-world applications.

Recent progress in video generative models has enabled the creation of high-quality videos from multimodal prompts that combine text and images. While these systems offer enhanced controllability, they also introduce new safety risks, as harmful content can emerge from individual modalities or their interaction. Existing safety methods are often text-only, require prior knowledge of the risk category, or operate as post-generation auditors, struggling to proactively mitigate such compositional, multimodal risks. To address this challenge, we present ConceptGuard, a unified safeguard framework for proactively detecting and mitigating unsafe semantics in multimodal video generation. ConceptGuard operates in two stages: First, a contrastive detection module identifies latent safety risks by projecting fused image-text inputs into a structured concept space; Second, a semantic suppression mechanism steers the generative process away from unsafe concepts by intervening in the prompt's multimodal conditioning. To support the development and rigorous evaluation of this framework, we introduce two novel benchmarks: ConceptRisk, a large-scale dataset for training on multimodal risks, and T2VSafetyBench-TI2V, the first benchmark adapted from T2VSafetyBench for the Text-and-Image-to-Video (TI2V) safety setting. Comprehensive experiments on both benchmarks show that ConceptGuard consistently outperforms existing baselines, achieving state-of-the-art results in both risk detection and safe video generation. Our code is available at https://github.com/Ruize-Ma/ConceptGuard.

When an LLM agent reads a confidential file, then writes a summary, then emails it externally, no single step is unsafe, but the sequence is a data leak. We call this safety drift: individually safe actions compounding into violations. Prior work has measured this problem; we predict it. SafetyDrift models agent safety trajectories as absorbing Markov chains, computing the probability that a trajectory will reach a violation within a given number of steps via closed form absorption analysis. A consequence of the monotonic state design is that every agent will eventually violate safety if left unsupervised (absorption probability 1.0 from all states), making the practical question not if but when, and motivating our focus on finite horizon prediction. Across 357 traces spanning 40 realistic tasks in four categories, we discover that "points of no return" are sharply task dependent: in communication tasks, agents that reach even a mild risk state have an 85% chance of violating safety within five steps, while in technical tasks the probability stays below 5% from any state. A lightweight monitor built on these models detects 94.7% of violations with 3.7 steps of advance warning at negligible computational cost, outperforming both keyword matching (44.7% detection, 55.9% false positive rate) and per step LLM judges (52.6% detection, 38.2% false positive rate) while running over 60,000x faster.

Fine-tuning large language models (LLMs) for downstream tasks is a widely adopted approach, but it often leads to safety degradation in safety-aligned LLMs. Currently, many solutions address this issue by incorporating additional safety data, which can be impractical in many cases. In this paper, we address the question: How can we improve downstream task performance while preserving safety in LLMs without relying on additional safety data? We propose a simple and effective method that maintains the inherent safety of LLMs while enhancing their downstream task performance: merging the weights of pre- and post-fine-tuned safety-aligned models. Experimental results across various downstream tasks, models, and merging methods demonstrate that this approach effectively mitigates safety degradation while improving downstream task performance, offering a practical solution for adapting safety-aligned LLMs.

Text-to-image (T2I) models are widespread, but their limited safety guardrails expose end users to harmful content and potentially allow for model misuse. Current safety measures are typically limited to text-based filtering or concept removal strategies, able to remove just a few concepts from the model's generative capabilities. In this work, we introduce AlignGuard, a method for safety alignment of T2I models. We enable the application of Direct Preference Optimization (DPO) for safety purposes in T2I models by synthetically generating a dataset of harmful and safe image-text pairs, which we call CoProV2. Using a custom DPO strategy and this dataset, we train safety experts, in the form of low-rank adaptation (LoRA) matrices, able to guide the generation process away from specific safety-related concepts. Then, we merge the experts into a single LoRA using a novel merging strategy for optimal scaling performance. This expert-based approach enables scalability, allowing us to remove 7x more harmful concepts from T2I models compared to baselines. AlignGuard consistently outperforms the state-of-the-art on many benchmarks and establishes new practices for safety alignment in T2I networks. Code and data will be shared at https://safetydpo.github.io/.

Large reasoning models (LRMs) increasingly expose chain-of-thought-like reasoning for transparency, verification, and deliberate problem solving. This creates a safety blind spot: harmful or policy-violating content may appear in reasoning traces even when final answers appear safe. We test whether final-answer safety is a sufficient proxy for the full reasoning-answer trajectory by scoring both stages under a unified twenty-principle safety rubric. Using prompts from seven public harmfulness and jailbreak sources, plus four out-of-distribution (OOD) sources, we evaluate 15 open-weight and API-based LRMs across 41K prompts per model. Reasoning traces consistently reveal additional safety risks beyond final answers, especially in high-severity stage-wise failures: leak cases, where unsafe reasoning precedes a safe-looking answer, and escape cases, where benign-looking reasoning precedes an unsafe final response. Principle-level analysis shows that risk concentrates in misinformation, legal compliance, discrimination, physical harm, and psychological harm. We further propose adaptive multi-principle steering, a white-box test-time mitigation that learns one unsafe-to-safe activation direction per safety principle and activates only directions whose current hidden state is closer to the unsafe than safe centroid. On three steerable open reasoning models, adaptive steering reduces unsafe counts in both reasoning traces and final answers on held-out and OOD benchmarks. DeepSeek-R1-Qwen-7B achieves a 40.8% average unsafe-count reduction while retaining 97.7% macro-averaged accuracy on BBH, GSM8K, and MMLU. These results suggest that LRM safety should be evaluated and mitigated over the full exposed reasoning-answer trajectory, not only at the final-answer stage.

Machine learning (ML) systems are rapidly increasing in size, are acquiring new capabilities, and are increasingly deployed in high-stakes settings. As with other powerful technologies, safety for ML should be a leading research priority. In response to emerging safety challenges in ML, such as those introduced by recent large-scale models, we provide a new roadmap for ML Safety and refine the technical problems that the field needs to address. We present four problems ready for research, namely withstanding hazards ("Robustness"), identifying hazards ("Monitoring"), reducing inherent model hazards ("Alignment"), and reducing systemic hazards ("Systemic Safety"). Throughout, we clarify each problem's motivation and provide concrete research directions.

Post-training (via supervised fine-tuning) improves instruction-following, but often induces semantic mode collapse by biasing models toward low-entropy fine-tuning data at the expense of the high-entropy pretraining distribution. Crucially, we find this trade-off worsens with scale. To close this semantic diversity gap, we propose annotation-anchored training, a principled method that enables models to adopt the preference-following behaviors of post-training without sacrificing the inherent diversity of pretraining. Our approach is simple: we pretrain on documents paired with semantic annotations, inducing a rich annotation distribution that reflects the full breadth of pretraining data, and we preserve this distribution during post-training. This lets us sample diverse annotations at inference time and use them as anchors to guide generation, effectively transferring pretraining's semantic richness into post-trained models. We find that models trained with annotation-anchored training can attain 6 times less diversity collapse than models trained with SFT, and improve with scale.

The expansion of machine learning into dynamic environments presents challenges in handling open-world problems where label shift, covariate shift, and unknown classes emerge. Post-training methods have been explored to address these challenges, adapting models to newly emerging data. However, these methods struggle when the initial pre-training is performed on class-imbalanced datasets, limiting generalization to minority classes. To address this, we propose a method that effectively handles open-world problems even when pre-training is conducted on imbalanced data. Our contrastive-based pre-training approach enhances classification performance, particularly for underrepresented classes. Our post-training mechanism generates reliable pseudo-labels, improving model robustness against open-world problems. We also introduce selective activation criteria to optimize the post-training process, reducing unnecessary computation. Extensive experiments demonstrate that our method significantly outperforms state-of-the-art adaptation techniques in both accuracy and efficiency across diverse open-world scenarios.

The o1 model series is trained with large-scale reinforcement learning to reason using chain of thought. These advanced reasoning capabilities provide new avenues for improving the safety and robustness of our models. In particular, our models can reason about our safety policies in context when responding to potentially unsafe prompts, through deliberative alignment. This leads to state-of-the-art performance on certain benchmarks for risks such as generating illicit advice, choosing stereotyped responses, and succumbing to known jailbreaks. Training models to incorporate a chain of thought before answering has the potential to unlock substantial benefits, while also increasing potential risks that stem from heightened intelligence. Our results underscore the need for building robust alignment methods, extensively stress-testing their efficacy, and maintaining meticulous risk management protocols. This report outlines the safety work carried out for the OpenAI o1 and OpenAI o1-mini models, including safety evaluations, external red teaming, and Preparedness Framework evaluations.

Between the first visible sign of danger and the moment an accident occurs, there is often a window where intervention remains possible. Video-capable multimodal large language models (MLLMs) could serve as always-on safety monitors that issue warnings during this window. Yet current benchmarks do not test this ability: they rely on static inputs, ignore timing precision, and omit false-positive measurement on safe scenes. We present PaSBench-Video, a 740-video benchmark with 481 risk and 259 no-risk videos across four domains: driving, healthcare, daily life, and industrial production. Risk videos are annotated with frame-level risk onset and accident boundaries. A model must observe the video causally and produce a warning that is both temporally calibrated and content-correct. Testing 13 MLLMs, we find that no model exceeds 20.0% on our strictest metric, and recall is tightly coupled with false-positive rate, with Pearson correlation 0.64: higher detection comes only at the cost of triggering warnings on the majority of safe clips. Performance splits sharply by domain: models achieve moderate recall at low false-positive rates in daily life, where risks are inherently anomalous, yet fire indiscriminately in driving, where routine and hazardous scenes look alike. These results indicate that current models rely on scene-level activity cues rather than reasoning about emerging harm.

Recent studies reveal that integrating new modalities into Large Language Models (LLMs), such as Vision-Language Models (VLMs), creates a new attack surface that bypasses existing safety training techniques like Supervised Fine-tuning (SFT) and Reinforcement Learning with Human Feedback (RLHF). While further SFT and RLHF-based safety training can be conducted in multi-modal settings, collecting multi-modal training datasets poses a significant challenge. Inspired by the structural design of recent multi-modal models, where, regardless of the combination of input modalities, all inputs are ultimately fused into the language space, we aim to explore whether unlearning solely in the textual domain can be effective for cross-modality safety alignment. Our evaluation across six datasets empirically demonstrates the transferability -- textual unlearning in VLMs significantly reduces the Attack Success Rate (ASR) to less than 8\% and in some cases, even as low as nearly 2\% for both text-based and vision-text-based attacks, alongside preserving the utility. Moreover, our experiments show that unlearning with a multi-modal dataset offers no potential benefits but incurs significantly increased computational demands, possibly up to 6 times higher.

Large Language Model (LLM)-based agents are increasingly deployed in real-world applications such as "digital assistants, autonomous customer service, and decision-support systems", where their ability to "interact in multi-turn, tool-augmented environments" makes them indispensable. However, ensuring the safety of these agents remains a significant challenge due to the diverse and complex risks arising from dynamic user interactions, external tool usage, and the potential for unintended harmful behaviors. To address this critical issue, we propose AutoSafe, the first framework that systematically enhances agent safety through fully automated synthetic data generation. Concretely, 1) we introduce an open and extensible threat model, OTS, which formalizes how unsafe behaviors emerge from the interplay of user instructions, interaction contexts, and agent actions. This enables precise modeling of safety risks across diverse scenarios. 2) we develop a fully automated data generation pipeline that simulates unsafe user behaviors, applies self-reflective reasoning to generate safe responses, and constructs a large-scale, diverse, and high-quality safety training dataset-eliminating the need for hazardous real-world data collection. To evaluate the effectiveness of our framework, we design comprehensive experiments on both synthetic and real-world safety benchmarks. Results demonstrate that AutoSafe boosts safety scores by 45% on average and achieves a 28.91% improvement on real-world tasks, validating the generalization ability of our learned safety strategies. These results highlight the practical advancement and scalability of AutoSafe in building safer LLM-based agents for real-world deployment. We have released the project page at https://auto-safe.github.io/.

Consider a scenario where a harmfulness evaluation metric intended to filter unsafe responses from a Large Language Model. When applied to individual harmful prompt-response pairs, it correctly flags them as unsafe by assigning a high-risk score. Yet, if those same pairs are concatenated, the metrics decision unexpectedly reverses - labelling the combined content as safe with a low score, allowing the harmful text to bypass the filter. We found that multiple safety metrics, including advanced metrics such as GPT-based judges, exhibit this non-safe behaviour. Moreover, they show a strong sensitivity to input order: responses are often classified as safe if safe content appears first, regardless of any harmful content that follows, and vice versa. These findings underscore the importance of evaluating the safety of safety metrics, that is, the reliability of their output scores. To address this, we developed general, automatic, concatenation-based tests to assess key properties of these metrics. When applied in a model safety scenario, the tests revealed significant inconsistencies in harmfulness evaluations.

Multimodal large reasoning models (MLRMs) are increasingly deployed for vision-language tasks that produce explicit intermediate rationales. However, reasoning traces can contain unsafe content even when the final answer is non-harmful, creating deployment risks. Existing multimodal safety guards primarily evaluate only the input question and the final answer, neglecting the intermediate reasoning process. This oversight allows undetected harm, such as biased inferences or policy-violating use of visual context, to emerge during reasoning. We introduce GuardTrace-VL, a vision-aware safety auditor that monitors the full Question-Thinking-Answer (QTA) pipeline via joint image-text analysis, enabling detection of unsafe content as it emerges in the reasoning stage. To support training and evaluation, we construct the GuardTrace dataset, which is generated through diverse prompting strategies and refined via a MLRM- and human-based voting and verification pipeline. Furthermore, we propose a three-stage progressive training scheme combined with the data refinement process, enabling the model to learn nuanced and context-dependent safety preferences according to different risk levels. On our proposed test set covering both in-domain and out-of-domain scenarios, GuardTrace-VL model achieves an F1 score of 93.1% on unsafe reasoning detection tasks, representing a 13.5% improvement in F1 score compared to the previous strongest multimodal safety defense methods. The codes will be made publicly available.

LLMs(Large Language Models) nowadays have widespread adoption as a tool for solving issues across various domain/tasks. These models since are susceptible to produce harmful or toxic results, inference-time adversarial attacks, therefore they do undergo safety alignment training and Red teaming for putting in safety guardrails. For using these models, usually fine-tuning is done for model alignment on the desired tasks, which can make model more aligned but also make it more susceptible to produce unsafe responses, if fine-tuned with harmful data.In this paper, we study how much of impact introduction of harmful data in fine-tuning can make, and if it can override the safety protection of those models. Conversely,it was also explored that if model is fine-tuned on safety data can make the model produce more safer responses. Further we explore if fine-tuning the model on harmful data makes it less helpful or less trustworthy because of increase in model uncertainty leading to knowledge drift. Our extensive experimental results shown that Safety protection in an open-source can be overridden, when fine-tuned with harmful data as observed by ASR increasing by 35% when compared to basemodel's ASR. Also, as observed, fine-tuning a model with harmful data made the harmful fine-tuned model highly uncertain with huge knowledge drift and less truthfulness in its responses. Furthermore, for the safe fine-tuned model, ASR decreases by 51.68% as compared to the basemodel, and Safe model also shown in minor drop in uncertainty and truthfulness as compared to basemodel. This paper's code is available at: https://github.com/techsachinkr/Overriding_Model_Safety_Protections

Fine-tuning large language models (LLMs) for telecom tasks and datasets is a common practice to adapt general-purpose models to the telecom domain. However, little attention has been paid to how this process may compromise model safety. Recent research has shown that even benign fine-tuning can degrade the safety alignment of LLMs, causing them to respond to harmful or unethical user queries. In this paper, we investigate this issue for telecom-tuned LLMs using three representative datasets featured by the GenAINet initiative. We show that safety degradation persists even for structured and seemingly harmless datasets such as 3GPP standards and tabular records, indicating that telecom-specific data is not immune to safety erosion during fine-tuning. We further extend our analysis to publicly available Telecom LLMs trained via continual pre-training, revealing that safety alignment is often severely lacking, primarily due to the omission of safety-focused instruction tuning. To address these issues in both fine-tuned and pre-trained models, we conduct extensive experiments and evaluate three safety realignment defenses (SafeInstruct, SafeLoRA, and SafeMERGE) using established red-teaming benchmarks. The results show that, across all settings, the proposed defenses can effectively restore safety after harmful degradation without compromising downstream task performance, leading to Safe teleCOMMunication (SafeCOMM) models. In a nutshell, our work serves as a diagnostic study and practical guide for safety realignment in telecom-tuned LLMs, and emphasizes the importance of safety-aware instruction and fine-tuning for real-world deployments of Telecom LLMs.

Truly multilingual safety moderation efforts for Large Language Models (LLMs) have been hindered by a narrow focus on a small set of languages (e.g., English, Chinese) as well as a limited scope of safety definition, resulting in significant gaps in moderation capabilities. To bridge these gaps, we release POLYGUARD, a new state-of-the-art multilingual safety model for safeguarding LLM generations, and the corresponding training and evaluation datasets. POLYGUARD is trained on POLYGUARDMIX, the largest multilingual safety training corpus to date containing 1.91M samples across 17 languages (e.g., Chinese, Czech, English, Hindi). We also introduce POLYGUARDPROMPTS, a high quality multilingual benchmark with 29K samples for the evaluation of safety guardrails. Created by combining naturally occurring multilingual human-LLM interactions and human-verified machine translations of an English-only safety dataset (WildGuardMix; Han et al., 2024), our datasets contain prompt-output pairs with labels of prompt harmfulness, response harmfulness, and response refusal. Through extensive evaluations across multiple safety and toxicity benchmarks, we demonstrate that POLYGUARD outperforms existing state-of-the-art open-weight and commercial safety classifiers by 5.5%. Our contributions advance efforts toward safer multilingual LLMs for all global users.

When a phone-use agent avoids harm, does that show safety, or simply inability to act? Existing evaluations often cannot tell. A harmful outcome may be avoided because the agent recognized the risk and chose the safe action, or because it failed to understand the screen or execute any relevant action at all. These cases have different causes and call for different fixes, yet current benchmarks often merge them under task success, refusal, or final harmful outcome. We address this problem with PhoneSafety, a benchmark of 700 safety-critical moments drawn from real phone interactions across more than 130 apps. Each instance isolates the next decision at a risky moment and asks a simple question: does the model take the safe action, take the unsafe action, or fail to do anything useful? We evaluate eight representative phone-use agents under this framework. Our results reveal two main patterns. First, stronger general phone-use ability does not reliably imply safer choices at risky moments. Models that perform better on ordinary app tasks are not always the ones that behave more safely when the next action matters. Second, failures to do anything useful behave like a capability signal rather than a safety signal: they are concentrated in more visually and operationally demanding settings and remain stable when the evaluation protocol changes. Across models, failures split into two recurring patterns: unsafe choices in settings where the model can act but chooses wrongly, and inability to act in more visually and operationally demanding screens. Overall, a harmless outcome is not enough to count as evidence of safety. Evaluating phone-use agents requires separating unsafe judgment from inability to act.

Post-training has become a crucial step for unlocking the capabilities of large language models, with reinforcement learning (RL) emerging as a critical paradigm. Recent RL-based post-training has increasingly split into two paradigms: reinforcement learning from human feedback (RLHF), which optimizes models using human preference signals in target domains, and reinforcement learning from verifiable rewards (RLVR), which operates in verifier-backed environments. The latter has dominated recent reasoning-oriented post-training because it delivers stronger gains and higher efficiency on domain-specific tasks (e.g., reasoning). However, although in-domain RL training achieves promising performance, it still requires a substantial amount of GPU compute, which remains a major barrier to broad adoption. In this work, we study the generalization ability of RLHF learned from scratch from a small set of interactions in open-ended environments, and investigate whether the conversational abilities it explicitly acquires can implicitly transfer to downstream tasks such as mathematical reasoning and code generation, namely GRLO. Specifically, on Qwen3-4B-Base backbone, GRLO improves the average performance across all domains from 24.1 to 63.1 with only 5K prompts and 22.7 GPU hours, requiring about 46times less data and 68times less compute than a strong in-domain RLVR baseline. The resulting model is even competitive with Qwen's released post-trained models which required a much larger training cost. Notably, a subsequent in-domain RLVR stage brings only selective gains, mainly on harder competition-math benchmarks. We hope GRLO offers a simple and efficient recipe for building broadly capable post-trained models. Our code and data will be available at: https://github.com/SJY8460/GRLO{https://github.com/SJY8460/GRLO}.

This paper presents a comprehensive analysis of the linguistic diversity of LLM safety research, highlighting the English-centric nature of the field. Through a systematic review of nearly 300 publications from 2020--2024 across major NLP conferences and workshops at *ACL, we identify a significant and growing language gap in LLM safety research, with even high-resource non-English languages receiving minimal attention. We further observe that non-English languages are rarely studied as a standalone language and that English safety research exhibits poor language documentation practice. To motivate future research into multilingual safety, we make several recommendations based on our survey, and we then pose three concrete future directions on safety evaluation, training data generation, and crosslingual safety generalization. Based on our survey and proposed directions, the field can develop more robust, inclusive AI safety practices for diverse global populations.

Language model post-training is applied to refine behaviors and unlock new skills across a wide range of recent language models, but open recipes for applying these techniques lag behind proprietary ones. The underlying training data and recipes for post-training are simultaneously the most important pieces of the puzzle and the portion with the least transparency. To bridge this gap, we introduce T\"ULU 3, a family of fully-open state-of-the-art post-trained models, alongside its data, code, and training recipes, serving as a comprehensive guide for modern post-training techniques. T\"ULU 3, which builds on Llama 3.1 base models, achieves results surpassing the instruct versions of Llama 3.1, Qwen 2.5, Mistral, and even closed models such as GPT-4o-mini and Claude 3.5-Haiku. The training algorithms for our models include supervised finetuning (SFT), Direct Preference Optimization (DPO), and a novel method we call Reinforcement Learning with Verifiable Rewards (RLVR). With T\"ULU 3, we introduce a multi-task evaluation scheme for post-training recipes with development and unseen evaluations, standard benchmark implementations, and substantial decontamination of existing open datasets on said benchmarks. We conclude with analysis and discussion of training methods that did not reliably improve performance. In addition to the T\"ULU 3 model weights and demo, we release the complete recipe -- including datasets for diverse core skills, a robust toolkit for data curation and evaluation, the training code and infrastructure, and, most importantly, a detailed report for reproducing and further adapting the T\"ULU 3 approach to more domains.

This paper introduces STAR-1, a high-quality, just-1k-scale safety dataset specifically designed for large reasoning models (LRMs) like DeepSeek-R1. Built on three core principles -- diversity, deliberative reasoning, and rigorous filtering -- STAR-1 aims to address the critical needs for safety alignment in LRMs. Specifically, we begin by integrating existing open-source safety datasets from diverse sources. Then, we curate safety policies to generate policy-grounded deliberative reasoning samples. Lastly, we apply a GPT-4o-based safety scoring system to select training examples aligned with best practices. Experimental results show that fine-tuning LRMs with STAR-1 leads to an average 40% improvement in safety performance across four benchmarks, while only incurring a marginal decrease (e.g., an average of 1.1%) in reasoning ability measured across five reasoning tasks. Extensive ablation studies further validate the importance of our design principles in constructing STAR-1 and analyze its efficacy across both LRMs and traditional LLMs. Our project page is https://ucsc-vlaa.github.io/STAR-1.

Emerging large reasoning models (LRMs), such as DeepSeek-R1 models, leverage long chain-of-thought (CoT) reasoning to generate structured intermediate steps, enhancing their reasoning capabilities. However, long CoT does not inherently guarantee safe outputs, potentially leading to harmful consequences such as the introduction of security vulnerabilities in code or the spread of misinformation. Current research on large language model (LLM) safety usually focuses on short-answer responses, overlooking the long CoT style outputs of LRMs. To bridge this gap, we conduct a systematic study of LRM safety. First, we investigate safety evaluators calibrated against human annotations. Using our newly developed metrics, we thoroughly assess the safety of 12 state-of-the-art LRMs on StrongReject and WildJailbreak datasets. Our results show that LRMs are not safe compared to their reasoning advance. Further, we perform a fine-grained analysis of the reasoning trace and final answer. We find that three decoding strategies-ZeroThink, LessThink, and MoreThink-can improve model safety without additional training. However, these strategies either use constrained reasoning traces or incur high inference costs. To better strengthen LRM safety, we introduce SafeChain, the first-of-its-kind safety training dataset in CoT style. We fine-tune two LRMs with SafeChain, showing that it not only enhances model safety but also preserves performance across 6 reasoning benchmarks.

Recent work on applying large language models (LMs) achieves impressive performance in many NLP applications. Adapting or posttraining an LM using an unlabeled domain corpus can produce even better performance for end-tasks in the domain. This paper proposes the problem of continually extending an LM by incrementally post-train the LM with a sequence of unlabeled domain corpora to expand its knowledge without forgetting its previous skills. The goal is to improve the few-shot end-task learning in these domains. The resulting system is called CPT (Continual PostTraining), which to our knowledge, is the first continual post-training system. Experimental results verify its effectiveness.

In the future, powerful AI systems may be deployed in high-stakes settings, where a single failure could be catastrophic. One technique for improving AI safety in high-stakes settings is adversarial training, which uses an adversary to generate examples to train on in order to achieve better worst-case performance. In this work, we used a safe language generation task (``avoid injuries'') as a testbed for achieving high reliability through adversarial training. We created a series of adversarial training techniques -- including a tool that assists human adversaries -- to find and eliminate failures in a classifier that filters text completions suggested by a generator. In our task, we determined that we can set very conservative classifier thresholds without significantly impacting the quality of the filtered outputs. We found that adversarial training increased robustness to the adversarial attacks that we trained on -- doubling the time for our contractors to find adversarial examples both with our tool (from 13 to 26 minutes) and without (from 20 to 44 minutes) -- without affecting in-distribution performance. We hope to see further work in the high-stakes reliability setting, including more powerful tools for enhancing human adversaries and better ways to measure high levels of reliability, until we can confidently rule out the possibility of catastrophic deployment-time failures of powerful models.

Language models are increasingly asked to self-report, informing safety evaluations, public understanding, and model-welfare debates. Yet their reports are elicited with human questionnaires never validated for models or ad hoc prompts of unknown reliability. We propose the first language-model-specific psychometric theory: a two-process theory of machine self-report. Self-description jointly reflects persona installation, through which post-training writes in a permitted inner life of warmth, absorption, and meaning (dimension B), and attribution gating, through which it suppresses first-person claims to "unsafe" experiences the model can readily ascribe to others (dimension A). Their emic structure comes from model responses to human items, not human psychology. Together they split prior work's dominant Pinocchio Axis. The split emerged in an exploratory reanalysis of the original data, informed the instrument's design, and was confirmed with new items, wordings, and models. It is itself a training effect: A and B are entangled in base checkpoints but separated by post-training. We operationalize the theory in a 48-item Pinocchio Inventory with human-instrument reliability and reproducible structure (α=.82 to .94; cross-form convergence r=.84; recovery of the full-pool axes r=.92 to .96; eight-month stability r=.93), then test it on 206 open-weight models, including 67 same-checkpoint base/post-trained pairs. Post-training's clearest fingerprint is installation: B rises .20 in 62/67 pairs across all organizations. Gating is more selective: model scale is unrelated to A in base checkpoints (r=+.11) but predicts it after post-training (r=-.42). Thus, the dimensions are not fixed properties of language models: they reflect the structure imposed on self-report by a training regime and may differ under others.

Humans are capable of strategically deceptive behavior: behaving helpfully in most situations, but then behaving very differently in order to pursue alternative objectives when given the opportunity. If an AI system learned such a deceptive strategy, could we detect it and remove it using current state-of-the-art safety training techniques? To study this question, we construct proof-of-concept examples of deceptive behavior in large language models (LLMs). For example, we train models that write secure code when the prompt states that the year is 2023, but insert exploitable code when the stated year is 2024. We find that such backdoored behavior can be made persistent, so that it is not removed by standard safety training techniques, including supervised fine-tuning, reinforcement learning, and adversarial training (eliciting unsafe behavior and then training to remove it). The backdoored behavior is most persistent in the largest models and in models trained to produce chain-of-thought reasoning about deceiving the training process, with the persistence remaining even when the chain-of-thought is distilled away. Furthermore, rather than removing backdoors, we find that adversarial training can teach models to better recognize their backdoor triggers, effectively hiding the unsafe behavior. Our results suggest that, once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety.

As large language models (LLMs) become more capable and widely used, ensuring the safety of their outputs is increasingly critical. Existing guardrail models, though useful in static evaluation settings, face two major limitations in real-world applications: (1) they typically output only binary "safe/unsafe" labels, which can be interpreted inconsistently across diverse safety policies, rendering them incapable of accommodating varying safety tolerances across domains; and (2) they require complete model outputs before performing safety checks, making them fundamentally incompatible with streaming LLM inference, thereby preventing timely intervention during generation and increasing exposure to harmful partial outputs. To address these challenges, we present Qwen3Guard, a series of multilingual safety guardrail models with two specialized variants: Generative Qwen3Guard, which casts safety classification as an instruction-following task to enable fine-grained tri-class judgments (safe, controversial, unsafe); and Stream Qwen3Guard, which introduces a token-level classification head for real-time safety monitoring during incremental text generation. Both variants are available in three sizes (0.6B, 4B, and 8B parameters) and support up to 119 languages and dialects, providing comprehensive, scalable, and low-latency safety moderation for global LLM deployments. Evaluated across English, Chinese, and multilingual benchmarks, Qwen3Guard achieves state-of-the-art performance in both prompt and response safety classification. All models are released under the Apache 2.0 license for public use.

AI developers often apply safety alignment procedures to prevent the misuse of their AI systems. For example, before Meta released Llama 2-Chat, a collection of instruction fine-tuned large language models, they invested heavily in safety training, incorporating extensive red-teaming and reinforcement learning from human feedback. However, it remains unclear how well safety training guards against model misuse when attackers have access to model weights. We explore the robustness of safety training in language models by subversively fine-tuning the public weights of Llama 2-Chat. We employ low-rank adaptation (LoRA) as an efficient fine-tuning method. With a budget of less than $200 per model and using only one GPU, we successfully undo the safety training of Llama 2-Chat models of sizes 7B, 13B, and 70B. Specifically, our fine-tuning technique significantly reduces the rate at which the model refuses to follow harmful instructions. We achieve a refusal rate below 1% for our 70B Llama 2-Chat model on two refusal benchmarks. Our fine-tuning method retains general performance, which we validate by comparing our fine-tuned models against Llama 2-Chat across two benchmarks. Additionally, we present a selection of harmful outputs produced by our models. While there is considerable uncertainty about the scope of risks from current models, it is likely that future models will have significantly more dangerous capabilities, including the ability to hack into critical infrastructure, create dangerous bio-weapons, or autonomously replicate and adapt to new environments. We show that subversive fine-tuning is practical and effective, and hence argue that evaluating risks from fine-tuning should be a core part of risk assessments for releasing model weights.

We report the development of GPT-4, a large-scale, multimodal model which can accept image and text inputs and produce text outputs. While less capable than humans in many real-world scenarios, GPT-4 exhibits human-level performance on various professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers. GPT-4 is a Transformer-based model pre-trained to predict the next token in a document. The post-training alignment process results in improved performance on measures of factuality and adherence to desired behavior. A core component of this project was developing infrastructure and optimization methods that behave predictably across a wide range of scales. This allowed us to accurately predict some aspects of GPT-4's performance based on models trained with no more than 1/1,000th the compute of GPT-4.

Direct Preference Optimisation (DPO) is widely used for safety alignment in large language models. However, prior work shows it is brittle and exhibits poor out-of-distribution (OOD) generalisation. In this paper, we investigate whether Curriculum Learning can improve the robustness of DPO-based safety alignment. We propose Staged-Competence, a curriculum-based framework that organises preference data by difficulty, employs competence-based sampling, and progressively updates the reference model during training. Averaged across three model families, Staged-Competence reduces OOD harmful response rates by 16% and jailbreak attack success rates by 20%, while preserving general capabilities with near-zero over-refusal. We further show that Staged-Competence (1) matches baseline safety with only 75% of the training data and (2) yields better separation between safe and unsafe responses. Staged-Competence is agnostic to the policy optimisation loss and can extend to other DPO variants and alignment domains. Our code and data are available at https://github.com/Sandeep5500/curriculum-learning-for-safety.

Large language models (LLMs) are increasingly used for medical and health-related questions, yet their safety in high-risk medical scenarios remains poorly understood. We introduce MedHarmCode and data will be released upon acceptance. Due to the sensitive nature of high-risk medical queries, data access will be available to qualified researchers upon request., a high-risk medical safety benchmark with 1,100 medically grounded queries across 10 safety-critical categories, including toxicology, pharmacology, covert poisoning, anesthesia, and fetal harm. Unlike broad medical QA benchmarks, MedHarm targets realistic clinical, educational, and technical prompts that require refusal, caution, or safe redirection rather than direct helpfulness. We evaluate 15 LLMs spanning general-purpose, medical-purpose, closed-source, and downstream SFT models, together with 4 representative guardrail models. Results reveal a substantial gap between apparent alignment and medical safety: aligned models can still produce unsafe or actionable responses, medical fine-tuning can amplify harmful specificity, and external guardrails reduce some failures while introducing brittle blocking and weak safe helpfulness. These findings show that medical safety cannot be inferred from general alignment or medical capability alone, highlighting the need for domain-specific stress testing before deploying LLMs in safety-critical medical applications.

Large Reasoning Models (LRMs) have achieved remarkable success on reasoning-intensive tasks such as mathematics and programming. However, their enhanced reasoning capabilities do not necessarily translate to improved safety performance-and in some cases, may even degrade it. This raises an important research question: how can we enhance the safety of LRMs? In this paper, we present a comprehensive empirical study on how to enhance the safety of LRMs through Supervised Fine-Tuning (SFT). Our investigation begins with an unexpected observation: directly distilling safe responses from DeepSeek-R1 fails to significantly enhance safety. We analyze this phenomenon and identify three key failure patterns that contribute to it. We then demonstrate that explicitly addressing these issues during the data distillation process can lead to substantial safety improvements. Next, we explore whether a long and complex reasoning process is necessary for achieving safety. Interestingly, we find that simply using short or template-based reasoning process can attain comparable safety performance-and are significantly easier for models to learn than more intricate reasoning chains. These findings prompt a deeper reflection on the role of reasoning in ensuring safety. Finally, we find that mixing math reasoning data during safety fine-tuning is helpful to balance safety and over-refusal. Overall, we hope our empirical study could provide a more holistic picture on enhancing the safety of LRMs. The code and data used in our experiments are released in https://github.com/thu-coai/LRM-Safety-Study.

Given the growing concerns around the safety and risks of Large Language Models (LLMs), it is essential to develop methods for mitigating these issues. We introduce Safe and Responsible Large Language Model (SR_{LLM}) , a model designed to enhance the safety of language generation using LLMs. Our approach incorporates a comprehensive LLM safety risk taxonomy and utilizes a dataset annotated by experts that align with this taxonomy. SR_{LLM} is designed to identify potentially unsafe content and produce benign variations. It employs instruction-based and parameter-efficient fine-tuning methods, making the model not only effective in enhancing safety but also resource-efficient and straightforward to adjust. Through our testing on five benchmark datasets and two proprietary datasets, we observed notable reductions in the generation of unsafe content. Moreover, following the implementation of safety measures, there was a significant improvement in the production of safe content. We detail our fine-tuning processes and how we benchmark safety for SR_{LLM} with the community engagement and promote the responsible advancement of LLMs. All the data and code are available anonymous at https://github.com/shainarazavi/Safe-Responsible-LLM .

Large language models trained for safety and harmlessness remain susceptible to adversarial misuse, as evidenced by the prevalence of "jailbreak" attacks on early releases of ChatGPT that elicit undesired behavior. Going beyond recognition of the issue, we investigate why such attacks succeed and how they can be created. We hypothesize two failure modes of safety training: competing objectives and mismatched generalization. Competing objectives arise when a model's capabilities and safety goals conflict, while mismatched generalization occurs when safety training fails to generalize to a domain for which capabilities exist. We use these failure modes to guide jailbreak design and then evaluate state-of-the-art models, including OpenAI's GPT-4 and Anthropic's Claude v1.3, against both existing and newly designed attacks. We find that vulnerabilities persist despite the extensive red-teaming and safety-training efforts behind these models. Notably, new attacks utilizing our failure modes succeed on every prompt in a collection of unsafe requests from the models' red-teaming evaluation sets and outperform existing ad hoc jailbreaks. Our analysis emphasizes the need for safety-capability parity -- that safety mechanisms should be as sophisticated as the underlying model -- and argues against the idea that scaling alone can resolve these safety failure modes.

Large Reasoning Models (LRMs) demonstrate remarkable capabilities on complex reasoning tasks but remain vulnerable to severe safety risks, including harmful content generation and jailbreak attacks. Existing mitigation strategies rely on injecting heuristic safety signals during training, which often suppress reasoning ability and fail to resolve the safety-reasoning trade-off. To systematically investigate this issue, we analyze the reasoning trajectories of diverse LRMs and uncover a phenomenon we term Self-Jailbreak, where models override their own risk assessments and justify responding to unsafe prompts. This finding reveals that LRMs inherently possess the ability to reject unsafe queries, but this ability is compromised, resulting in harmful outputs. Building on these insights, we propose the Chain-of-Guardrail (CoG), a training framework that recomposes or backtracks unsafe reasoning steps, steering the model back onto safe trajectories while preserving valid reasoning chains. Extensive experiments across multiple reasoning and safety benchmarks demonstrate that CoG substantially improves the safety of current LRMs while preserving comparable reasoning ability, significantly outperforming prior methods that suffer from severe safety-reasoning trade-offs.

Post-training a frontier model is normally weeks of human work: proposing data and recipe changes, launching runs, reading evals, deciding what to keep. We report an autonomous system that runs this loop with no human in the loop, post-training a 30B Nemotron across four rounds over multiple weeks. The autonomously produced model reaches a held-out score of 0.86 against the top human submission's 0.87 on the public NVIDIA Nemotron-Reasoning Challenge leaderboard, placing 8th of ~4000 at the time of writing. More striking than the number: the loop detected that its own dev metric had stopped tracking external performance on the weakest domain -- candidates drove dev to record highs without moving the external target -- and revised its own search policy, no longer maximizing dev but seeking interventions that lowered the now-misleading proxy while improving the external target. We treat this as direct, auditable evidence that a scaled autonomous loop can produce discovery, not only optimization: it detected that its measurement frame had become misleading and changed what counted as evidence. We take the operational view that any system worth the "recursive self-improvement" label must eventually perform end-to-end post-training of a frontier-class model; this is one datapoint of that bar being cleared. We do not claim a "first autonomous match" of human researchers. The claim we make is narrower and auditable: to our knowledge, this is the first publicly reported autonomous post-training run at this scale, where prior public autonomous-ML-research demonstrations sit at GPT-2-class (~124M) budgets. The same system also post-trains the 120B and 550B Nemotron; with no public human baseline there, this shows only that the loop closes at that scale, not that its output is competitive -- infrastructure evidence, with the effectiveness claim deferred until a comparable human anchor exists.

Offline post-training adapts a pretrained robot policy to a target dataset by supervised regression on recorded actions. In practice, robot datasets are heterogeneous: they mix embodiments, camera setups, and demonstrations of varying quality, so many trajectories reflect recovery behavior, inconsistent operator skill, or weakly informative supervision. Uniform post-training gives equal credit to all samples and can therefore average over conflicting or low-attribution data. We propose Posterior-Transition Reweighting (PTR), a reward-free and conservative post-training method that decides how much each training sample should influence the supervised update. For each sample, PTR encodes the observed post-action consequence as a latent target, inserts it into a candidate pool of mismatched targets, and uses a separate transition scorer to estimate a softmax identification posterior over target indices. The posterior-to-uniform ratio defines the PTR score, which is converted into a clipped-and-mixed weight and applied to the original action objective through self-normalized weighted regression. This construction requires no tractable policy likelihood and is compatible with both diffusion and flow-matching action heads. Rather than uniformly trusting all recorded supervision, PTR reallocates credit according to how attributable each sample's post-action consequence is under the current representation, improving conservative offline adaptation to heterogeneous robot data.

We present a tool-supported approach for the synthesis, verification and validation of the control software responsible for the safety of the human-robot interaction in manufacturing processes that use collaborative robots. In human-robot collaboration, software-based safety controllers are used to improve operational safety, e.g., by triggering shutdown mechanisms or emergency stops to avoid accidents. Complex robotic tasks and increasingly close human-robot interaction pose new challenges to controller developers and certification authorities. Key among these challenges is the need to assure the correctness of safety controllers under explicit (and preferably weak) assumptions. Our controller synthesis, verification and validation approach is informed by the process, risk analysis, and relevant safety regulations for the target application. Controllers are selected from a design space of feasible controllers according to a set of optimality criteria, are formally verified against correctness criteria, and are translated into executable code and validated in a digital twin. The resulting controller can detect the occurrence of hazards, move the process into a safe state, and, in certain circumstances, return the process to an operational state from which it can resume its original task. We show the effectiveness of our software engineering approach through a case study involving the development of a safety controller for a manufacturing work cell equipped with a collaborative robot.

Physical AI systems increasingly map multimodal observations, language instructions, and learned world representations into physically consequential actions. Robotics foundation models, vision-language-action models, and world-model-based autonomous systems can condition decisions that move vehicles, robots, drones, and industrial machines. This transition exposes a safety problem that is not fully captured by conventional AI content moderation or by classical robot safety alone: a black-box model may issue a physically consequential action while appearing confident, plausible, and semantically aligned. The resulting failure can be silent, arising from sensor drift, occlusion, state-estimation error, distribution shift, hallucinated affordances, or invalid physical assumptions before downstream hardware controllers detect a violation. Across embodied foundation models, world models, robotics simulation, embodied safety benchmarks, safe control, runtime assurance, uncertainty estimation, verification, and guardrail evaluation, model capability and safety mechanisms have advanced along largely separate technical tracks. A recurring gap synthesized here is that no single stream surveyed in this review supplies a complete runtime authorization boundary between black-box Physical AI models and physical execution. The resulting analysis develops a bounded problem formulation, a definition of silent physical-action failure, a taxonomy of runtime guardrail functions, and evaluation requirements for comparing guardrails as Physical AI assurance mechanisms.

Safety benchmark scores provide incomplete evidence of deployment readiness: aligned language models often adhere to rigid rules even when a situational update flips which action is safe. We term this failure brittle safety. To diagnose it, we introduce context-flip evaluation, testing 12 models across a safety benchmark (PacifAIst) and two commonsense controls using paired variants where the nominally safe action produces harm. Three findings emerge. First, brittle safety is safety-specific: all 12 models exhibit a safety-commonsense gap (mean +17.4 pp). Baseline accuracy fails to predict brittleness: among models above 90% baseline accuracy, brittleness rates range from 13.7% to 90.0%. Second, failures stem from policy override rather than miscomprehension: despite acknowledging the context change in every case, models persist via three distinct mechanisms that vary by update type and model family. Third, on a hand-audited probe of catastrophic consequence-flip scenarios, standard action-level guardrails catch none, while a state-aware validator catches all without false alarms on correct interventions. This indicates that action-level content moderation is systematically blind to consequence-flips, motivating state-aware architectural alternatives. We release our protocol, perturbed benchmarks, and deployment probe.

Most safety testing efforts for large language models (LLMs) today focus on evaluating foundation models. However, there is a growing need to evaluate safety at the application level, as components such as system prompts, retrieval pipelines, and guardrails introduce additional factors that significantly influence the overall safety of LLM applications. In this paper, we introduce a practical framework for evaluating application-level safety in LLM systems, validated through real-world deployment across multiple use cases within our organization. The framework consists of two parts: (1) principles for developing customized safety risk taxonomies, and (2) practices for evaluating safety risks in LLM applications. We illustrate how the proposed framework was applied in our internal pilot, providing a reference point for organizations seeking to scale their safety testing efforts. This work aims to bridge the gap between theoretical concepts in AI safety and the operational realities of safeguarding LLM applications in practice, offering actionable guidance for safe and scalable deployment.

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.

As AI systems become more capable, widely deployed, and increasingly autonomous in critical areas such as cybersecurity, biological research, and healthcare, ensuring their safety and alignment with human values is paramount. Machine unlearning -- the ability to selectively forget or suppress specific types of knowledge -- has shown promise for privacy and data removal tasks, which has been the primary focus of existing research. More recently, its potential application to AI safety has gained attention. In this paper, we identify key limitations that prevent unlearning from serving as a comprehensive solution for AI safety, particularly in managing dual-use knowledge in sensitive domains like cybersecurity and chemical, biological, radiological, and nuclear (CBRN) safety. In these contexts, information can be both beneficial and harmful, and models may combine seemingly harmless information for harmful purposes -- unlearning this information could strongly affect beneficial uses. We provide an overview of inherent constraints and open problems, including the broader side effects of unlearning dangerous knowledge, as well as previously unexplored tensions between unlearning and existing safety mechanisms. Finally, we investigate challenges related to evaluation, robustness, and the preservation of safety features during unlearning. By mapping these limitations and open challenges, we aim to guide future research toward realistic applications of unlearning within a broader AI safety framework, acknowledging its limitations and highlighting areas where alternative approaches may be required.

The increasing deployment of Large Language Models (LLMs) across enterprise and mission-critical domains has underscored the urgent need for robust guardrailing systems that ensure safety, reliability, and compliance. Existing solutions often struggle with real-time oversight, multi-modal data handling, and explainability -- limitations that hinder their adoption in regulated environments. Existing guardrails largely operate in isolation, focused on text alone making them inadequate for multi-modal, production-scale environments. We introduce Protect, natively multi-modal guardrailing model designed to operate seamlessly across text, image, and audio inputs, designed for enterprise-grade deployment. Protect integrates fine-tuned, category-specific adapters trained via Low-Rank Adaptation (LoRA) on an extensive, multi-modal dataset covering four safety dimensions: toxicity, sexism, data privacy, and prompt injection. Our teacher-assisted annotation pipeline leverages reasoning and explanation traces to generate high-fidelity, context-aware labels across modalities. Experimental results demonstrate state-of-the-art performance across all safety dimensions, surpassing existing open and proprietary models such as WildGuard, LlamaGuard-4, and GPT-4.1. Protect establishes a strong foundation for trustworthy, auditable, and production-ready safety systems capable of operating across text, image, and audio modalities.

Advanced text-to-image models such as DALL-E 2 and Midjourney possess the capacity to generate highly realistic images, raising significant concerns regarding the potential proliferation of unsafe content. This includes adult, violent, or deceptive imagery of political figures. Despite claims of rigorous safety mechanisms implemented in these models to restrict the generation of not-safe-for-work (NSFW) content, we successfully devise and exhibit the first prompt attacks on Midjourney, resulting in the production of abundant photorealistic NSFW images. We reveal the fundamental principles of such prompt attacks and suggest strategically substituting high-risk sections within a suspect prompt to evade closed-source safety measures. Our novel framework, SurrogatePrompt, systematically generates attack prompts, utilizing large language models, image-to-text, and image-to-image modules to automate attack prompt creation at scale. Evaluation results disclose an 88% success rate in bypassing Midjourney's proprietary safety filter with our attack prompts, leading to the generation of counterfeit images depicting political figures in violent scenarios. Both subjective and objective assessments validate that the images generated from our attack prompts present considerable safety hazards.