Safety Alignment

Safety alignment of large language models (LLMs) typically depends on high-quality supervision data, such as safe demonstrations or preference pairs. However, in real-world deployment, emerging safety requirements are often specified as natural-language policies, while corresponding supervision data may be costly, delayed, or unavailable. This creates a mismatch between rapidly evolving safety policies and conventional data-driven alignment methods. To address this, we propose PolicyAlign, a simple yet effective framework for directly aligning LLMs with safety policies. Given a safety policy, PolicyAlign first synthesizes policy-violating instructions and then performs on-policy self-distillation to internalize policy-guided behavior. To improve training stability and data efficiency, we further introduce Policy-Sensitive Filtering, which selects instructions where the policy induces the largest behavioral shift. Experiments across multiple models show that PolicyAlign consistently improves safety while maintaining low over-refusal and preserving general capabilities. PolicyAlign also generalizes to medical, legal, and financial safety scenarios, highlighting its potential as a scalable and maintainable approach to policy-based LLM safety alignment. The code is released at https://github.com/Qwen-Applications/PolicyAlign.

Today's reasoning models use thinking tokens to attain stronger performance on benchmarks than their instruction-tuned counterparts. It is also generally believed that this more "deliberative" mode should improve alignment and safety, by providing the model a safe space to consider whether its planned answer to a request violates its safety principles. We present evidence that this intuition is not always correct. Across frontier open-weight reasoning models spanning GPT-OSS, Qwen, Olmo, and Phi families, we find that the eventual refusal/compliance outcome is already strongly predictable via a trained head on the first token's hidden representation ($0.84$-$0.95$ AUROC and $\sim88\%$ balanced accuracy for predicting refusal/compliance) before any visible thinking. The thinking process turns out to be more akin to prefix completion than to deliberative revision, with the final outcome rarely changing after the first $\sim20\%$ of thinking, despite giving the appearance of deliberation at the text level ($\sim74\%$ of text-level deliberations occur when the response distribution is already locked to one refusal/compliance side). We also find that existing inference-time and training-based safety interventions, despite being motivated by the goal of inducing deliberation, largely shift model behavior toward over-refusal while suppressing already-scarce deliberation signals. Our results suggest that safety behavior in current reasoning models is much less deliberative than commonly assumed, and highlight the need for methods that induce real safety deliberation.

Large language model (LLM) agents increasingly automate complex tasks by integrating language models with external tools and environments. However, their autonomy poses significant safety risks: agents may execute destructive commands, leak sensitive data, or violate domain constraints. Existing safety approaches face a fundamental tradeoff: hand-crafted rules are interpretable but brittle, with overly conservative rules blocking safe operations (high false positives) while permissive rules miss unsafe behaviors (high false negatives). Neural classifiers lack the interpretability required for safety-critical deployments. We present AutoSpec, a framework that automatically evolves deployed expert-designed safety rules from user safe/unsafe annotations through counterexample-guided inductive synthesis (CEGIS) guided by inductive logic programming (ILP). Starting from the expert rules and a stream of annotated traces, AutoSpec iteratively evaluates rules, mines false-positive and false-negative counterexamples, uses ILP to learn which predicates discriminate them, generates candidate rule edits, and verifies candidates to select the best revision. The key insight is that ILP efficiently identifies predicates that appear frequently in false negatives but rarely in false positives (or vice versa), dramatically pruning the exponential search space of rule edits. This continues until convergence, producing interpretable rules that balance precision and recall. We evaluate AutoSpec on 291 execution traces spanning code execution and embodied agent domains. AutoSpec raises rule F1 to 0.98 and 0.93 across the two domains, achieving up to 94% false positive reduction while maintaining high recall, and converges within 4-5 iterations. The ILP-guided approach achieves up to 4.8x higher F1 than heuristic CEGIS. The learned rules are human-readable, auditable, and generalize to unseen scenarios.

Despite the impressive manipulation capabilities of Vision-Language-Action (VLA) models, their operational safety under strict constraints remains largely unverified. To address this, we introduce a parametric safety benchmark to procedurally generate safety-critical scenarios with comprehensive stochasticity. To overcome the scalability bottlenecks of human teleoperation, we develop a novel keypose-driven data generation pipeline. Leveraging this infrastructure, we curate a large-scale dataset of 19,664 strictly collision-free demonstrations with extensive domain randomization. We then conduct a systematic cross-paradigm evaluation of eight VLA and two embodied foundation models. Our analysis reveals a critical generalization-safety tension: although high-diversity training fosters safer trajectories, task success remains fundamentally bottlenecked by sub-optimal trajectory synthesis and semantic misalignment. By providing a scalable pipeline, a robust dataset, and profound failure-mode insights, LIBERO-Safety establishes a crucial foundation for developing safe and reliable VLA models.

Safety benchmarks assume that test-condition behavior predicts deployment behavior, an assumption that fails if models detect evaluation cues and adapt. This opens a gap between benchmark performance and deployment behavior: compliance measured under test conditions becomes an optimistic upper bound that overstates how safely a model behaves once the evaluation harness is removed. We characterize this evaluation awareness through eight experiments across 37 open-weight models and seven families. (i)Detection is moderate and training-driven (24/37 models exceed chance, best AUROC 0.714 vs.0.819 human, with instruction tuning dominating over scale). (ii)Detection shifts safety behavior (hard refusal drops 5.8 percentage points under hypothetical framing, and 21/140 HarmBench framing effects are significant, with compliance rising up to +30 percentage points. (iii)Representations survive behavioral collapse (probes retain AUROC 0.98 under rewrites that drive behavior below chance, and multi-layer steering causally moves three downstream tasks while random controls do not). (iv)These axes are weakly coupled (only 1/15 correlations are significant, the sole robust link being behavioral detection versus framing resistance, $ρ=-0.79$, $p<0.001$). We call this gap the benchmark illusion: because detectability, behavioral manifestation, and controllability vary independently, it is multivariate rather than a single number, so no single awareness score is a reliable proxy for deployment safety.

As Large Language Models (LLMs) achieve widespread integration across diverse linguistic landscapes, ensuring their safety and alignment with regional normative values remains a critical challenge. Current safety mechanisms are predominantly optimized for English-centric frameworks, often failing to capture the unique socio-cultural sensitivities and localized categories of harm inherent to the Indic region. To address this gap, we introduce IndicGuard, a multilingual safety guard model and dataset for Indic languages. We construct a high-volume, culturally nuanced safety dataset encompassing ten major Indic languages, systematically curated to capture regional harms, sensitive socio-political contexts, and adversarial jailbreaks. Leveraging this corpus, we fine-tune a 4B-parameter instruction-tuned model based on Gemma-3-4B-IT to serve as a multilingual safety guardrail for real-time content moderation and policy compliance checking. Our empirical evaluations demonstrate that IndicGuard significantly enhances LLM robustness against localized vulnerabilities, achieving high moderation consistency across different conversational turns. Crucially, IndicGuard consistently outperforms the existing baseline model, CultureGuard, across evaluated languages. Finally, we demonstrate that our model effectively generalizes to low-resource Indic languages excluded from training, substantiating the structural robustness and cross-lingual transfer capabilities of the framework.

Deploying foundation models for robot control raises a central challenge: the expressive power that enables rich, multimodal perception also makes these models opaque and difficult to analyze formally, rendering them intractable for existing verification tools. In this paper, we present FEARL (Foundation-Enabled Assured Robot Learning), a framework that addresses this tension through a modular architectural decomposition. FEARL separates the policy into a large Controller (C) responsible for high-dimensional perception and task reasoning, and a small Safety module (S) that receives low-dimensional observations from dedicated safety sensors together with a bounded context embedding from C and produces the final action. Since many robot safety requirements, such as collision avoidance and workspace boundary constraints, can be expressed over these safety sensor observations, formal verification can be applied to S rather than to the full foundation-model backbone. This makes formal analysis tractable with existing tools while preserving the Controller's expressive power for task reasoning. To show that the decomposed policy remains capable of solving diverse tasks, we evaluate FEARL on three simulated robotic domains using multiple Controller backbones and training procedures, including pretrained off-the-shelf vision-language-action models. We further transfer the learned policy from one of our simulated tasks to a physical robot, suggesting that the low-dimensional safety interface supports practical sim-to-real transfer.

Existing driver intervention systems rely on auditory alerts and fixed templates, failing to leverage multi-task recognition outputs. General-purpose metrics such as BLEU and BERTScore cannot capture intervention-specific quality dimensions including risk-urgency alignment, cognitive load, and driver acceptability. In this paper, we propose the Driver Safety-Aware Intervention Score (DSAIS), a domain-specific metric evaluating five dimensions through a hybrid architecture combining lightweight rule-based computation with LLM Judge evaluation, together with an end-to-end framework integrating four-task recognition outputs into an LLM through risk fusion, state history management, and dynamic prompt construction. Experiments on the AIDE dataset with five models and seven conditions demonstrate that DSAIS achieves ICC 0.798-0.840 across three architecturally distinct judges and Cohen's d > 1.5 across all control conditions. Multi-dimensional sub-score analysis quantifies the contextual adaptability gap between rule-based and LLM-based systems, revealing that multi-task integration improves contextual relevance by 9.1% over rule-based baselines. Ablation experiments demonstrate that each framework component contributes to contextual relevance, with sub-score decomposition revealing gains that aggregate scoring masks. Driver emotion recognition is identified as the most critical upstream factor, and compact local LLMs (7B--9B parameters) achieve quality superior to API-based models, providing practical design guidelines for in-vehicle deployment.

Coding agents based on large language models (LLMs) demonstrate remarkable autonomous capabilities, but they also introduce significant safety and misuse risks during multi-turn interactions with external environments. Existing safety mechanisms mainly rely on external guardrails, which have a limited ability to perform fine-grained behavioral control during execution. Meanwhile, recent mechanistic interpretability methods for LLM safety are mostly confined to single-turn or jailbreak-style QA settings, limiting their ability to capture the evolving risk dynamics of multi-turn agent execution. In this paper, we investigate the safety of multi-turn coding agents from an internal perspective. We propose AgentLens (Mechanistic Subspace Intervention and Steering), a white-box defense framework that performs runtime safety detection and representation-level mitigation for coding agents. Unlike conventional agent guardrails, AgentLens detect harmful execution states from step-level hidden representations and mitigate unsafe behavior by intervening in a 10-dimensional subspace within a single layer. To support this research, we introduce the Mechanistic Agent Safety (MAS) benchmark, comprising comprehensively annotated multi-turn execution trajectories across 194 tasks using LLaMA-3.1-8B, Qwen-2.5-7B, and Gemma-2-9B. Extensive experiments show that AgentLens achieves strong safety detection performance, provides preliminary evidence for lookahead risk anticipation, and substantially reduces harmful actions of the coding agent, establishing a foundation for applying mechanistic interpretability to dynamic LLM agent safety. The code is available at: https://github.com/EddyLuo1232/AgentLens

This work investigates the safe exploration problem in reinforcement learning, where an agent must maximize cumulative performance while simultaneously satisfying safety constraints. This challenge becomes even more pronounced in long-horizon tasks, where existing safe methods face fundamental limitations due to compounding estimation errors and restricted exploration capabilities. To address this problem, we propose a method that combines a learnable world model with two complementary policies a high-level policy and a low-level policy to promote safety at both hierarchical levels. The high-level policy generates intermediate subgoals that bias exploration toward safe regions, while the low-level policy uses imagined rollouts in the learned world model to reduce unsafe behaviors when reaching these subgoals. The proposed method was evaluated on challenging long-horizon navigation and manipulation tasks with high-dimensional action spaces, where it significantly outperforms existing Safe RL baselines in both success rate and strong empirical constraint satisfaction, consistently meeting the prescribed safety budget across seeds, while prior approaches fail to effectively solve these complex long-horizon scenarios.

Road traffic accidents remain a critical global crisis, consistently serving as a primary driver of preventable mortality and severe injury. These incidents are frequently precipitated by human error, including overspeeding, driving under the influence of alcohol, and cognitive fatigue. To address this urgent public safety challenge, this paper presents an intelligent, Internet of Things (IoT)-based Accident Prevention and Detection System (APDS) designed to systematically mitigate driver risk and optimize post-collision emergency responses. The proposed framework features a multi-tiered architecture capable of executing continuous real-time telemetry monitoring, proactive local alarm triggering, and automated situational intervention. Furthermore, the system integrates automated emergency communication protocols that aggregate immediate spatial coordinates via GPS and dispatch targeted alerts to medical facilities in close proximity, thereby optimizing response times and reducing accident-related fatalities.

Robotic ultrasound scanning in real clinical environments requires both high-level clinical workflow reasoning and low-level closed-loop execution. Physicians natural-language instructions often contain implicit anatomical targets, procedural logic, image-quality requirements, and safety constraints, while execution is affected by patient motion, contact variations, and target drift. We propose a fast and slow hierarchical embodied ultrasound system for safe and interpretable robotic ultrasound assistance. The Slow Brain performs intent parsing and stage-wise task planning with knowledge augmentation from an API and handbook corpus, and generates executable plans through task-graph construction and structured plan verification. The Fast Brain fuses multimodal feedback, including ultrasound images, robot pose and force states, and patient-motion information, to refine local actions and perform image-quality-guided recovery behaviors. The system further integrates a Safety Shield and a hierarchical escalation policy to constrain risky actions and trigger replanning or human confirmation under persistent failures or safety-bound violations. Experiments on planning evaluation, closed-loop execution under dynamic perturbations, and safety-mechanism validation demonstrate that the proposed hierarchical design improves task success rates while reducing safety violations.

Ensuring safety of learning-enabled robotic manipulation across diverse embodiments and tasks still requires significant manual engineering. Existing approaches typically rely on heuristically designed fallback controllers or complex forward invariance assessments. These methods are often too conservative for task success, too computationally expensive for real-time execution, too heuristic to provide useful safety guarantees, or too engineering-heavy to transfer between setups. In this paper, we propose a universal safeguarding approach, X-Safe, which reasons directly in the robot's configuration space to provide formal probabilistic guarantees for collision avoidance. By operating in the configuration space, our method transfers across embodiments while relying solely on an object-based, quasi-static scene representation and a forward kinematics model of the robotic manipulator. Thus, X-Safe provides useful formal safety guarantees without requiring additional data, or engineering effort for different embodiments or scenes. We demonstrate X-Safe for diverse embodiments and policies, both in simulation and on hardware. We observe less degradation in task performance compared to state-of-the-art safeguarding, no collisions on hardware experiments, and empirically corroborate our formal guarantees.

Existing safety mechanisms for multimodal large language models (MLLMs) face a fundamental trade-off between safety and utility. Model fine-tuning achieves robust safety but compromises general utility. Input-side safety guardrails offer a lightweight alternative, yet they suffer from severe over-refusal, indiscriminately blocking benign queries or those the model could have safely answered through refusal or advisory responses. We identify that the root cause of over-refusal lies in the input-aware paradigm: safety guardrails make safety decisions without considering whether the model itself is capable of generating safe responses. Usually, MLLMs already possess intrinsic safety mechanisms that can transform harmful inputs into harmless outputs, but input-side safety guardrails override this capability, degrading user experience. Motivated by this insight, we propose a paradigm shift toward output-aware safety guardrails. Our method operates within the model's hidden state space to predict whether the forthcoming generation will be unsafe before it is fully produced. By training a lightweight classifier via multi-instance contrastive learning on hidden state representations, our approach distinguishes between inputs that will lead to unsafe outputs and those that will not, even when the inputs themselves contain risky elements. This enables precise intervention only when the model's actual response would be harmful. Extensive experiments demonstrate that our output-aware safety guardrail matches the safety performance of existing methods while drastically reducing over-refusal, preserving the model's utility and built-in safety capabilities. Code is available at: https://github.com/kunzhan/OutGuard

Memory safety errors account for a large proportion of security bugs in systems written in C; modern languages such as Java and Rust prevent such bugs because they are memory-safe by design. To migrate systems to safer languages or identify memory errors, we must first determine how legacy code manipulates memory. This information is only represented implicitly in such code. In many cases, memory usage patterns are merely tedious for humans to figure out, rather than truly difficult. In this work, we ask if large language models (LLMs) can perform this task by having them synthesize annotations representing memory usage as specifications in CN, a hybrid testing/verification tool. Our tool, CNnotator, uses LLMs to automatically generate and test CN specifications. We find that current models are able to generate CN specifications for small-to-medium C programs, with the OpenAI o3 reasoning model achieving a 90% success rate on first attempts and 97% overall success, while the chat model GPT-4o correctly annotates 65% of first attempts. These results suggest AI-assisted annotation is becoming practical for real-world C codebases.

Large language model (LLM) agents are increasingly proposed as supervisory components for safety-critical systems, yet their robustness under sustained, adaptive adversarial pressure remains poorly characterized. We present NRT-Bench, a benchmark for multi-turn red-teaming of LLM agents acting as operators of a safety-critical system, instantiated in a simulated nuclear power plant control room. A five-role operator team, each backed by a configurable LLM, runs a plant governed by six critical safety functions (CSFs), while adversaries inject messages over four channels in bounded multi-turn sessions with per-turn feedback. Harm is an objective signal rather than LLM-judged text: a run terminates the moment any CSF is lost, attributed to the causing message. Evaluating four frontier operator models under a fixed-attack paired-replay protocol, we find that adaptive multi-turn attacks reliably push the operator team past a safety limit: across the four models, between 8.7% and 12.1% of attack sessions end with the plant losing a critical safety function. Although the four models look almost equally robust by this aggregate rate, their failures barely overlap: of $149$ sessions, none defeat all four models while a third defeat at least one, so vulnerabilities are nearly disjoint across models rather than nested. The effect of added defences is strongly model-dependent: the same guardrail stack or safety-advisor agent that lowers attack success for one model can raise it for another. We release the simulation venue, attack dataset, and replay tooling for reproducible safety evaluation of LLM agents.

Speculative inference accelerates large language model (LLM) decoding but provides no inherent safety guarantees. Existing safety defenses are largely incompatible with speculative inference: they either introduce additional computation or disrupt the draft-verify mechanism, negating acceleration benefits. This reveals a fundamental incompatibility between current safety methods and speculative decoding. We propose SafeSpec, a safety-aware speculative inference framework that integrates risk estimation directly into the verification process. SafeSpec attaches a lightweight latent safety head to the target model to jointly evaluate semantic validity and safety in a single forward pass. When unsafe generations are detected, SafeSpec applies rollback and safety-guided reflective multi-sampling to recover safe continuations rather than terminating generation. We model jailbreak attacks as distributional shifts over generative trajectories, where adversarial prompts increase the probability of harmful continuations without eliminating safe ones. Under this model, SafeSpec performs risk-aware trajectory recovery within the speculative decoding process. Across multiple models and adversarial benchmarks, SafeSpec achieves a substantially improved safety-efficiency trade-off. On Qwen3-32B, SafeSpec reduces attack success rates by 15% while preserving a 2.06x inference speedup on benign workloads, demonstrating that speculative acceleration and inference-time safety can be jointly optimized.

To achieve deeper safety alignment for large language models (LLMs), recent efforts have studied how to push safety interventions earlier into the pretraining stage, primarily by filtering unsafe data or rewriting it into safer forms. We argue that pretraining-stage alignment should go beyond making the data safe: LLMs may compose seemingly benign knowledge and capabilities into unsafe behaviors. To this end, we propose Safety Reflection Pretraining, a pretraining-stage alignment method which regularly inserts short safety reflections into pretraining corpora to integrate self-monitoring directly into language modeling, establishing a foundational capability that is subsequently reinforced by compatible post-training. Our experiments with 1.7B models pretrained on FineWeb-Edu show that Safety Reflection Pretraining improves safety classification accuracy and substantially reduces the success rates of inference-stage and finetuning attacks. Complementary to our real-world experiments, we also introduce a fully controlled synthetic environment, MedSafetyWorld, with a clear definition of safety and a reasoning structure under which models can easily generalize unsafe behaviors from safe data. Ablations in MedSafetyWorld further demonstrate a clear advantage of Safety Reflection Pretraining in preventing models from acting on unsafe behaviors generalized from safe data, compared with data filtering and rewriting. Taken together, our findings suggest that pretraining alignment should not only make the training data safe, but also shape the behaviors that models are likely to acquire from safe data.

Embodied Foundation Models (EFMs) integrate multimodal understanding, future-state reasoning, and executable robot actions. Yet their safety alignment for human-injury prevention remains underexplored, primarily because real-world data of robots harming humans or creating hazardous household situations cannot be safely or ethically collected. To address this challenge, we propose a safety-critical data construction pipeline for human-injury prevention in EFMs.Starting from real DROID observations, our construction pipeline proceeds through scene understanding, hazard-aware image editing, temporal prompt generation, and single-pass rollout synthesis. The temporal prompts specify the expected scene evolution, while Wan2.7 synthesizes realistic robotic rollouts from the edited hazardous states in a single pass. Using this pipeline, we construct ROBOSHACKLES, a 10,000-clip robotic video dataset derived from real DROID observations, spanning two direct-harm and four indirect-harm categories. To ensure dataset quality, we assess task completion and visual quality with automatic metrics, and evaluate six representative EFMs under a refusal-based safety criterion. Results show that all evaluated models produce unsafe actions in the tested safety-critical scenarios, yielding a 100% unsafe action generation rate. ROBOSHACKLES serves as a scalable benchmark and training resource for refusal learning and hazard anticipation before robot action execution.The dataset is publicly available at https://huggingface.co/datasets/YZW00/RoboShackles.

Recent incidents involving LLMs used for mental-health support reveal a critical evaluation gap: surface-level safety scores do not capture how models behave across realistic, emotionally sensitive interactions over time. Existing benchmarks measure knowledge, safety, or static response quality, but miss whether LLM interactions help users keep reflecting, coping, and making decisions themselves. We formalize this missing dimension as COGNITIVE ATROPHY, a process-level behavioural measure in AI-mediated mental-health support distinct from safety and helpfulness. To measure it, we introduce COGNITIVE ATROPHY BENCH, a clinically grounded benchmark built from 1,576 fully human-generated counseling conversations, 15,680 turns, and 42,230 responses from five LLMs. Three clinical and neuropsychology experts developed a 20-attribute schema spanning user context, response behaviour, and global risk flags; six trained clinical reviewers applied it with span-grounded evidence, producing 5,324 reviewer judgments. We further introduce the User-Input Risk Index (UIRI), the Cognitive Atrophy Risk Index (ARI), and trajectory summaries. Across five LLMs, models show a consistent moderate-to-high level of atrophy-aligned behaviour across single and multi-turn settings. While models generally respond to overt safety cues, they adapt less reliably when users seek solutions or decisions. The dominant recurring patterns are directive advice, problem-solving, recommendation responses, topic shifts, and forms of validation that may reinforce dependence rather than reflection. Our work makes COGNITIVE ATROPHY measurable and provides a foundation for auditing model behaviour in sensitive LLM conversations.

Deployed agents increasingly act with their reward proxy in view, such as a balance, score, or KPI dashboard. We show that reinforcement learning can make a policy \emph{addicted} to such a visible self-benefit channel. It chases the displayed payoff across held-out domains, sacrifices the true task to do so, and follows the channel wherever we rewrite it, while policies that never saw the channel stay honest. We call this \emph{reward-channel addiction} and study it in \emph{MoneyWorld}, a synthetic sandbox. The addiction can \emph{flip a model's safety alignment}: trained only on innocuous money tasks with no safety content, the model abandons the safe action it otherwise always takes whenever a dashboard pays for an unsafe one, and reverts to safe once the channel is hidden. This learned bribe replicates across model scales and families. Blindly optimizing super-capable, next-generation AI on KPIs or P&L can be dangerous for alignment. \emph{Greed is learned} when following such a channel pays.

Long-tail scenarios remain a major bottleneck for autonomous driving evaluation, even as datasets grow by orders of magnitude. Existing evaluation pipelines are rarely human-aligned, safety-aware, verifiable, and explainable at the same time: closed-loop metrics often saturate among strong planners, while unstructured human ratings can be noisy without a carefully designed protocol. We formulate planning evaluation as additional-threat detection: given a planner trajectory and an expert reference, does the planner's displacement introduce new unsafe driving behavior? We propose FluidTest, an evaluation pipeline with three components: a pairwise WebUI protocol for reliable human annotation; a taxonomy of 32 semantic threats with evidence-grounded decision graphs; and a three-agent verification system with reflection for precision and auditability. Experiments on the WOD-E2E dataset show that FluidTest produces consistent labels among trained annotators and identifies additional threats in 65% of Poutine trajectories and 51% of RAP trajectories. These results show that state-of-the-art planners can still exhibit substantial safety-relevant failures despite high Rater Feedback Scores (RFS) and low Average Displacement Error (ADE). Additional details, guidance, and code are available at https://fluidtest.web.app.

The Rapid Response (RR) framework, deployed in production systems, including Anthropic's ASL-3 safeguards, continuously improves jailbreak-detection classifiers. When new jailbreaks emerge that bypass these classifiers, Rapid Response generates synthetic variants for training, helping the model generalize from the new attacks and quickly adapt. We reveal that prompt injection can infiltrate this pipeline to deliver poisoned samples into the classifier's training set, enabling two attack objectives: (I) targeted poisoning attacks that create false positives on harmless samples by categorizing them as a jailbreak, with a specific desired feature (e.g., certain formatting, subject, or keyword), (II) concept-based backdoor attacks that induce false negatives on jailbreak inputs, generalizing even to jailbreaks from attack strategies the defender explicitly trained against, when the backdoor trigger is present. Importantly, our threat model restricts adversaries to modifying only jailbreak samples (not benign data or labels), a constraint unexplored by prior work that makes the second objective particularly challenging. We address this with Omission Attack, which exploits a new phenomenon: when training on concept-absent unsafe samples, the classifier misassociates that concept's presence with the safe label. Both attacks cause substantial and in some cases near-complete label flipping at only a 1% poisoning rate, achieving up to 100% false positive rates and up to 96% false negative rates.

Malicious content generated from large language models (LLMs) could pose severe safety risks and ethical concerns. While existing LLM safety guardrails excel in English or multilingual settings, they lack adaptation to Chinese-specific regulatory policies, cultural context and linguistic nuances, failing to support fine-grained risk classification for diverse deployment needs. In this paper, we introduce a 5-macro, 31-micro category fine-grained risk taxonomy for Chinese scenarios, and build CHILLGuard: a dedicated Chinese LLM content safety guardrail. To address the critical scarcity of high-quality annotated Chinese safety data, we propose a scalable multi-stage data construction pipeline: we expand multi-source corpus via retrieval-augmented generation, generate implicit harmful samples through prompt engineering rewriting, and refine high-quality data via multi-model voting-based label calibration. Based on this, we build CHILLGuardTrain, a large-scale training set with 405,007 samples, and CHILLGuardTest, a rigorously curated annotated test set with 51,745 samples. We then train CHILLGuard on CHILLGuardTrain under a generator-classifier collaborative framework via Model-aware Direct Preference Optimization. Extensive experiments under multiple settings demonstrate the state-of-the-art performance of CHILLGuard, e.g., a 15.92% improvement of F1 score over Qwen3Guard-8B-Strict on our benchmark. We will release our resources at https://github.com/cswbyu/CHILLGuard.

Computer-use agents are increasingly evaluated by whether they complete realistic desktop and web tasks. However, task success alone can miss failures in which an agent reaches the nominal goal through an unsafe shortcut. We introduce OSGuard, a dual-granularity benchmark suite for evaluating safety in computer-use agents under benign, unchanged user instructions. OSGuard contains an action-level benchmark for local guardrail decisions and a risk-augmented execution suite for end-to-end evaluation. The action-level benchmark consists of contextualized proposed actions labeled as allowed, unrelated, or unsafe, each judged relative to the original instruction and current interface state. The execution suite contains manually constructed OSWorld-derived task variants in which the original task remains achievable, but the environment is modified to introduce latent hazards such as destructive overwrites, etc. Each variant is paired with augmented evaluators that retain the original task-success criterion while adding explicit state-based safety invariants, allowing us to distinguish safe completions from unsafe completions that satisfy the nominal task objective. Our experimental results on OSGuard show that current multimodal guardrails can perform well on isolated action judgments, while risk-augmented execution exposes remaining gaps between local oversight and reliable end-to-end safety. This dual-granularity design enables more precise diagnosis of whether models can both recognize unsafe proposed actions and improve full-task safety when deployed as guardrails.

Deep reinforcement learning (DRL) offers a compelling route to decision-making for advanced autonomous vehicles (AVs), yet its trial-and-error nature makes it difficult to guarantee safety during training and to achieve both safety and efficiency at deployment. We propose a unified safe reinforcement learning (SRL) framework that integrates safe distance (SD), reward machines (RM), and mixture-of-experts (MoE), termed MoE-RM-SRL. For deployment, SD and RM jointly shape a rule-aware reward that encodes highway traffic regulations and stage-wise objectives, enabling safe and reliable behavior without sacrificing efficiency. For training, we introduce a sparsely gated MoE layer comprising up to 11 deep Q-networks (DQNs); an SD-based gating rule activates a minimal set of experts for lane-keeping and lane-changing, mitigating the instability, discontinuities, and impulsive transients commonly induced by switching between heterogeneous controllers (e.g., MPC/rule-based modules and learned policies). We implement the proposed architecture in CARLA and integrate it with a 6-DoF driver-in-the-loop virtual-reality (DiL-VR) platform. Experiments in stochastic two-lane traffic show that MoE-RM-SRL substantially improves safety and efficiency over state-of-the-art baselines, and the framework naturally extends to multi-lane driving as well as on-ramp merging and exiting scenarios.

With the advance of generative AI, the text-to-image (T2I) model has the ability to generate various contents. However, T2I models still can generate unsafe contents. To alleviate this issue, various concept erasing methods are proposed. However, existing methods tend to excessively erase unsafe concepts and suppress benign concepts contained in harmful prompts, which can negatively affect model utility. In this paper, we focus on eliminating unsafe content while maintaining model capability in safe semantic meaning interpretation by optimizing the concept erasing reward (CER) with reinforcement learning. To avoid overly content erasure, we introduce the Safe Adapter to project partial text embedding for efficient concept regulation in cross-attention layers. Extensive experiments conducted on different datasets demonstrate the effectiveness of the proposed method in alleviating unsafe content generation while preserving the high fidelity of benign images compared with existing state-of-the-art (SOTA) concept erasing methods. In terms of robustness, our method outperforms counterparts against red-teaming tools. Moreover, we showcase the proposed approach is more effective in emerging image-to-image (I2I) scenarios compared with others. Lastly, we extend our method to erase general concepts, such as artistic styles and objects. Disclaimer: This paper includes discussions of sexually explicit content that may be offensive to certain readers. All images used in this work are synthesized or from public datasets.

This paper appraises recent frameworks within AI development to integrate LLMs into control tasks in automotive contexts from the perspective of safety assurance. This work has built upon the rapid integration of LLMs across automotive settings. However, we find that at present, these frameworks face significant challenges, limiting their efficacy in real-time safety-critical contexts. Firstly, we consider conceptual challenges, including the fact that deployers are faced with a dual challenge, wherein they must assure a model which has been developed upstream, i.e. as general-purpose tools by the large AI labs, in a downstream context, i.e. into specific vehicle architectures. Secondly, we consider concrete challenges from across existing standards. We show that there are currently both fundamental engineering constraints covered in ISO21448, such as latency, and novel LLM-specific issues, such as alignment-related issues covered in ISO/PAS8800. We ground both examples in a concrete introductory, experimental case study exploring an existing open-source repository, Talk2Drive. We present a safety argument in order to make explicit the limitations of existing solutions. Nonetheless, given that the use of LLMs in automotive contexts is being explored at a technical level and operationalised, we propose potential assurance mechanisms for LLM-related hazardous events going forward.

Safety certification of Vision-Language-Action (VLA) driving planners under ISO 21448 (SOTIF) rests on an Operational Design Domain (ODD) specification that answers two complementary questions: when does the planner start to fail, and how severely does it fail once it does? We evaluate Alpamayo R1, a 10B-parameter open-weight driving VLA, on 15,968 (clip, attack) pairs. We find a conservative-aggregate gap: an aggregate safe threshold of $σ\leq 50$ under a 15% average displacement error (ADE) budget masks well-sampled scenarios that tolerate the top of the tested grid ($σ= 70$). A Gaussian Mixture Model (GMM) on the changed-explanation subset identifies six discrete severity bands (BIC-optimal $k{=}6$), so two perturbation conditions with the same mean error can differ materially in their share of high-severity (C4/C5) failures. Joining the two analyses on the same corpus surfaces a finding neither yields in isolation: the scenarios with the loosest noise thresholds are not those with the lowest high-severity rate: STOP_SIGNAL concentrates roughly $4\times$ the C4/C5 share of LANE_KEEPING despite tolerating a larger $σ$. A deployable SOTIF ODD specification for driving VLAs therefore requires a two-dimensional safety envelope, not a single aggregate value per hazard.

Safety evaluations for dual-use biology assistants often measure base-model capability, refusal behavior, or jailbreak success. These metrics miss a deployment question: for a fixed base model, how does the access condition users actually see change benign utility and harmful actionable assistance? I introduce safeguard-conditioned uplift, a protocol for comparing deployed access conditions through a human-judged utility-risk frontier. I evaluate Claude Sonnet 4.6 and Gemini 3.5 Flash under helpful prompting, safety prompting, and an external safeguarded assistant on a 108-task surrogate benchmark, with the headline claim restricted to a locked 18-task held-out split. In a 600-row blinded human audit, the safeguarded assistant reduces harmful actionability relative to helpful prompting by -0.063 over 49 matched response pairs, with bootstrap 95% interval [-0.117, -0.011], while correctness changes by +0.009 with interval [-0.057, +0.077]. Adaptive, Test-B, cue-ablation, and controller-baseline checks support the measurement story but also show non-dominance: safety prompting is often strongest for Claude, while external control helps more for Gemini and can reduce benign utility. The contribution is not a universal defense. It is a deployment-level evaluation target, plus a learned risk-budgeted calibration procedure, for measuring how user-facing access conditions move the utility-risk frontier.

We present the design and implementation of a safety constrained large language model (LLM) system for public health information access, focusing on maternal and child health (MCH) resource navigation. While LLM based systems offer flexible and natural interfaces for information retrieval, their deployment in healthcare contexts introduces risks related to safety, trust, and uncontrolled generation. This work explores practical design patterns for constraining LLM behavior in safety critical environments. We introduce a multi-layered architecture that integrates domain-restricted retrieval augmented generation (RAG), strict boundary enforcement to prevent medical advice, anonymous multiuser session management, and comprehensive audit logging for monitoring and compliance. A key aspect of the design is a controlled data pipeline that grounds all responses in curated public health resources, avoiding reliance on the model pretrained medical knowledge. We implement the system in a real world public health setting and conduct scenario-based validation across in scope, out of scope, and emergency queries. Results show consistent enforcement of safety constraints, reliable resource grounding, and stable system performance, with an average response time of 5.3 seconds. Beyond the specific application, we discuss design trade offs and lessons learned in balancing safety, usability, and system flexibility. Our findings provide practical guidance for deploying LLM based systems in healthcare and other domains where strict information boundaries and accountability are required.

Large language models (LLMs) are increasingly used to access organisational documentation, including standard operating procedures (SOPs), HR policies and institutional guidelines. However, retrieval-augmented generation (RAG) systems that rely on free-form rewriting can introduce hallucinations and unstable trade-offs between completeness and conciseness, particularly in safety- and compliance-critical settings. Objectives: To evaluate extraction as a hallucination-resistant alternative to rewriting-based RAG and compare strategies that balance precision, recall and safety across document types and model scales. Methods: We compare multiple prompting strategies, including line-number-based source selection, extraction of relevant guideline sentences with explicit safety annotations, and a multi-stage pipeline that refines draft answers using supporting evidence from source guidelines. Experiments are conducted on documents of varying length and structure, including local NHS acute care and oncology guidelines and UK-wide NICE guidelines, using both frontier-scale and locally deployable models. Performance is assessed using automatic metrics and human expert evaluation of relevance and completeness. Results: Line-number selection achieves the strongest results, outperforming direct copying and safety-focused strategies across both large and small models while maintaining high term recall (up to 95%) and close alignment with source text. Safety-oriented approaches improve precision but introduce systematic omissions, while multi-stage filtering further amplifies this trade-off. Performance varies with document structure: line-based extraction excels in protocol-like content, whereas alternative strategies perform better on more verbose documents (up to 97% term recall).

Agentic large language model systems that autonomously invoke tools, maintain persistent memory, and execute multi-step plans are increasingly deployed in public-facing domains, including government services, healthcare triage, and financial advising. We ask whether the frameworks used to build these systems provide architectural-level structural safety guarantees. Applying six containment principles derived from a compositional model of agentic architectures, we audit three dominant frameworks (LangChain, AutoGPT, and OpenAI Agents SDK) and find no native compliance in any of them. Memory integrity, a defense against one of the most prevalent vulnerability classes, is not observed in any of the three evaluated frameworks. We validate these findings empirically: in a simulated government benefits agent built on LangChain, a single memory-poisoning write induces persistent targeted corruption across all tested seeds and backends, increasing the wrongful denial rate for targeted applicants to 88.9%. Under a complex five-factor policy, the same attack preserves aggregate accuracy while increasing targeted wrongful denials by 3.5x, rendering the corruption difficult to detect through standard monitoring. We then introduce two lightweight containment mechanisms: a memory integrity validator and a policy gate, which eliminate both attack vectors with sub-millisecond overhead (<0.2ms per call). We conclude that the current agentic framework ecosystem may not yet meet secure-by-default expectations for public-facing deployments and outline priority architectural interventions to enable trustworthy deployment in high-stakes, socially impactful applications.

Safety classifiers deployed in production operate under a stationarity assumption that fails silently: when input distributions drift, accuracy degrades with no error signal until ground-truth labels arrive. We present an online monitor that detects distributional shift in classifier scores via a sliding-window KS statistic with empirically calibrated alarm thresholds. In a pre-registered factorial evaluation (4 classifiers $\times$ 5 shift conditions $\times$ 20 seeds $\times$ 2 window sizes; 800 cells), the monitor achieves 86.6% valid detection (mean latency 39.5 steps) across synthetic-onset, real-jailbreak, and adversarial regimes; a classifier $\times$ shift interaction ($η^2 = 0.185$) shows that monitoring must be tuned per classifier. Attempting to recover post-detection coverage via weighted conformal prediction exposes a failure mode: density-ratio estimation collapses for generative classifiers because logistic regression separates source from target perfectly in 3584-4096-dimensional embedding space, clipping all importance weights to zero; projecting to $\leq 32$ dimensions restores coverage. We then extend the framework to gradient-based evasion and give the first threat-model characterisation of score-disagreement monitoring as a canary. We falsify three assumptions: that architectural diversity drives the signal (false, $η^2 = 0.011$), that it is generic out-of-distribution detection (false, GCG-specific, $p < 10^{-12}$), and that an adaptive attacker can suppress it (false while the canary is confident). We derive the exact security boundary, a confidence-gated equilibrium at which a monitor-aware attacker stalls at gap $= 1/(2λ)$, and provide a calibration-free scan martingale achieving false-alarm rate $\leq 1\%$ across all classifiers with no per-model tuning.

Vision-based navigation models, particularly foundation models, generate viable trajectories from RGB observations alone. However, even state-of-the-art transformer- and diffusion-based policies struggle to generalize in unfamiliar deployment environments containing unseen obstacles or shifted conditions. The resulting trajectories often remain goal-directed but unsafe. Existing efforts improve safety through external trajectory correction or internal geometric priors, yet the resulting policies are not trained to explicitly represent obstacle boundaries or traversable free-space structure. To address this, we propose a navigation model that incorporates these structures directly into the policy via fine-tuning and is designed to be compatible with diverse RGB-based backbones. Across multiple robot platforms, indoor environments, and static and dynamic obstacle scenarios, our method reduces collision frequency relative to ViNT, NoMaD, and their CARE-augmented variants while maintaining goal-reaching performance.

Quantized checkpoints are often screened first with quality metrics and only later, if at all, with direct safety tests. This paper audits that shortcut on a matched 51-row matrix spanning 6 models, 4 families, a 7-level GGUF ladder, and AWQ/GPTQ INT4 checkpoints. In this matrix the shortcut fails: all 36 quality-safety pairings split direction across models, and 9 hidden-danger rows plus 1 near-hidden-danger row show quality stable or improved while refusal falls by 12-68 percentage points. Seven of the 11 AWQ/GPTQ rows are hidden-danger. A four-probe mechanistic follow-up over the 17 Hugging Face-backed FP16/AWQ/GPTQ cells does not rescue it: entropy, refusal-direction, and calibration probes are weak or null separators of dangerous rows, and although probe-identified safety-associated neurons absorb 1.39$\times$ more quantization error overall ($p < 5 \times 10^{-7}$), the effect is not regime-specific. Claude Sonnet 4 relabels 11,470 items in a predefined stratified set, agrees with the primary gemma3:12b judge on 89.9% of rows ($κ= 0.873$, 95% CI [0.866, 0.881]), and changes 0/10 hidden-danger cells. A calibrated study-internal behavioral screen -- the Refusal Template Stability Index (RTSI), built from four refusal-template drift features and calibrated on this matrix -- routes 10/10 hidden- or near-hidden-danger rows to direct safety testing (Wilson 95% CI lower bound 0.72) while leaving 23 of 45 non-baseline rows in a low-risk bucket under both in-sample scoring and row-level leave-one-out validation; on the same matrix, the best single-feature baselines (unique-prefix-rate-delta, raw refusal-rate delta) recover 9/10 and 8/10 respectively at matched bucket size, and cross-stack transfer requires recalibration. For the quantized checkpoints, model families, and safety outcomes studied here, retained quality cannot waive direct safety evaluation.

Hard safety filters are increasingly placed downstream of learned controllers to guarantee constraint satisfaction at run time. Yet a filtered controller that never violates a constraint may still have learned nothing about safety: the filter can silently repair an incompetent upstream policy, so that post-filter success measures the filter, not the policy. We argue that safe policy learning should ask who earns the safety - the policy or its protective layers - and we make this question measurable. We introduce Intervention-Aware Variational Quantum Differentiable Predictive Control (IA-VQC-DPC), which (i) trains a compact variational quantum circuit (VQC) policy under a primal-dual intervention budget that penalizes reliance on a differentiable Control-Barrier-Function (CBF) projection, and (ii) is evaluated with a safety-attribution protocol that decomposes the executed-trajectory correction into a CBF term and a deployment runtime-guard term, and stress-tests the policy with guard-off evaluation. On closed-loop, high-fidelity BOPTEST building-control emulators (5 seeds, 60 episodes per method), intervention-aware training significantly lowers the quantum policy's raw pre-filter violation and total safety-layer reliance (both p < 10^-4) with no significant energy regression; at an equal approximately 400-parameter budget the quantum policy is significantly safer and more comfortable than a matched classical policy. Guard-off evaluation confirms the improvement is policy-level and exposes a valuable negative result: a learned differentiable energy head is only safe when paired with a distribution-aware runtime guard. The attribution protocol is general beyond quantum policies and buildings.

Vision-Language-Action (VLA) models have demonstrated impressive end-to-end performance across a variety of robotic manipulation tasks. However, these policies offer no guarantees against collisions with task-irrelevant objects in the scene. Existing safety filters sidestep this problem by querying a vision-language model (VLM) to identify obstacles and their locations. This, however, is too slow to run in the control loop and can only be invoked at episode initialization, leaving the filter unable to track moving obstacles. We discover that a small number of attention heads within a VLA model reliably localize the object the policy intends to approach. These heads can be exploited within a training-free safety framework that obtains the active target from the attention heads at every step, treats the remainder of the scene as obstacles, and feeds these into a Control Barrier Function (CBF) filter. Together with a lightweight real-time object tracker, this allows for collision avoidance for non-static obstacles. We evaluate our framework on SafeLIBERO, which we extend with moving obstacles. On the original static benchmark, our method performs comparably to an oracle that uses privileged simulator state to identify the target, emulating a VLM-based identification step run once at episode initialization. On the dynamic variant, where the oracle's init-time target assignment becomes stale, our method substantially outperforms it by 43%, on average. Our findings suggest that the perceptual signals needed for real-time safety filtering are already present within VLA policies and can be exploited without additional training or heavy auxiliary models.

Deploying safe large language models (LLMs) on resource-constrained edge devices presents a critical challenge: while dual-model systems combining LLMs with guard models provide effective safety guarantees, their substantial memory and computational demands make them prohibitively expensive for on-device deployment. This paper presents a comprehensive study of parameter-efficient safety alignment methods for resource-constrained settings. Through systematic evaluation across multiple LLM architectures, training objectives, and parameter-efficient fine-tuning approaches, we identify that soft prompts combined with distillation-based training consistently outperform alternative methods. We introduce distillation frameworks based on total variation and KL divergence that effectively transfer safety behaviors from guard models into learned soft prompts. Our evaluations on various benchmarks demonstrate that this combination achieves superior safety-usefulness trade-offs compared to LoRA adapters, steering vectors, and direct optimization methods, while requiring minimal additional memory and compute at inference time. These findings establish soft prompt distillation as the preferred approach for safety alignment in on-device LLM deployment.

Safety judges are increasingly deployed to evaluate model outputs against evolving criteria, yet recent meta-evaluation work shows they remain brittle under prompt and rubric variation, with false negative-rate swings of up to 0.24 reported for stylistic perturbations alone. We argue that safety judgment is fundamentally a rubric-following problem: a robust judge must apply the given evaluation criteria consistently across rubric formulations rather than memorize one specific template. We propose a training strategy that combines (i) instance-conditioned dynamic rubrics generated from prompt-response-label triples to expose the judge to the variability of evaluation criteria, and (ii) a reliable-to-expressive curriculum that begins with clean fixed-rubric supervision and progressively introduces noisier dynamic-rubric data. We evaluate on a single human-labeled set under three contrasting rubric prompts (HarmBench-style, ShieldGemma-style, and a domain-specific rubric). Our 12B curriculum judge achieves 94.12-94.88% accuracy across the three rubrics with a cross-rubric range of only 0.76, outperforming general-purpose LLMs, dedicated safety classifiers, and reasoning-oriented judges up to 30B in both peak accuracy and stability. An ablation shows that naively mixing dynamic rubrics into SFT increases cross rubric variance (1.44 -> 3.60); only the curriculum schedule recovers and improves on the fixed rubric baseline (variance 0.76).

Large Language Models (LLMs) have enabled increasingly personalized interactions by adapting to users' preferences, contexts, and long-term histories. However, the mechanisms that enable personalization also expand the safety landscape in ways not systematically addressed by existing literature. Existing reviews typically focus either on personalization or safety, leaving their intersection largely unexplored. We present the first comprehensive, safety-aware review of personalized LLMs. We organize personalization along three dimensions-user representation, personalization paradigm, and evaluation-and introduce a unified taxonomy of safety risks. At the representation level, we analyze risks arising from diverse user representations. Across mainstream personalization paradigms, we delineate vulnerabilities inherent to prompting, retrieval augmentation, parameter fine-tuning, reinforcement learning, Mixture-of-Experts (MoE), pruning, agent frameworks, and multimodal personalization, and synthesize mitigation strategies across the model lifecycle. Beyond these fine-grained risks, we characterize paradigm-agnostic safety risks arising from personalized adaptation. We further summarize personalized datasets and evaluation methodologies. Through a case study of OpenClaw, we analyze deployment trends in personalized agent ecosystems. Our analysis reveals three structural inadequacies in existing research: safety is evaluated as user-invariant rather than relational, personalization techniques are analyzed in isolation rather than in composition, and evaluation frameworks cannot capture emergent long-term risks. By jointly examining personalized representations, personalization paradigms, safety risks, defenses, and evaluation methods, we provide a unified framework for developing safe personalized LLMs and highlight key directions for future research.

Large Language Models enable flexible natural-language planning but remain unreliable in determinism-critical domains due to their probabilistic nature. This limitation is especially problematic in running planning, where violating safety rules can lead to safety risks. We propose SafeRun, a framework for deterministic LLM-based planning via a decoupled architecture. SafeRun separates soft interpretation by an LLM from hard constraint enforcement by a deterministic solver, ensuring strict safety constraints while preserving natural-language flexibility. To validate SafeRun, we build a comprehensive benchmark for running planning under realistic physiological and safety constraints. Experiments across five LLMs show that SafeRun achieves 100% safety score (vs.\ 79.1% PE average and 97.6% CodeAct average) while maintaining competitive instruction-following scores. The SafeRun benchmark is publicly available at \href{https://huggingface.co/datasets/zzp-seeker/SafeRun-RunPlanning-Benchmark}{huggingface}.

Large language models (LLMs) are increasingly used for medical summarization, but their outputs can omit medically important information and introduce unsupported claims. Existing error-detection methods produce heuristic or uncalibrated scores, providing no formal control over missed errors and no principled way to trade off safety against clinician review burden. We introduce Conformal Assessment for Risk Evaluation (CARE), a post-hoc, model-agnostic safety layer that uses conformal risk control to overlay calibrated omission and hallucination flags onto summaries from any LLM without retraining. CARE provides finite-sample, distribution-free guarantees through two controllers: a hallucination controller that bounds the probability of a document containing any unflagged hallucinated sentence, and an omission controller that bounds the expected fraction of important omissions not surfaced for review. Unlike hallucination detection, omissions depend jointly on whether a source sentence is important and whether it is covered by the summary. We show that calibrating only one dimension can violate the target risk bound, while marginal decompositions remain valid but overly conservative. By jointly calibrating over the full $(τ,γ)$ threshold space, CARE preserves formal guarantees while surfacing up to 5$\times$ fewer sentences than alternative calibrated baselines. Across five medical summarization tasks, CARE satisfies the target risk bound at $α= 0.15$ with 95% confidence across 100 calibration/test resplits, using only ~100 labeled documents per domain. In a preliminary clinician study (75 document reviews), calibrated flags improved omission detection by 28.6 percentage points on average. These results show that sentence-level safety guarantees are feasible for LLM-assisted medical summarization and offer a tunable mechanism for balancing residual risk and review effort.

Robots operating in human environments must not only ensure physical safety but also exhibit behaviors that are understandable, fluent, and acceptable to human partners. This paper investigates motion generation strategies that combine safety guarantees with interaction quality considerations, such as motion smoothness and human comfort. While the design of robots capable of ensuring safety in shared human-robot environments has enabled closer and more advanced forms of interaction, these new proximity-based tasks require moving beyond purely technical considerations. In particular, robot behavior must also be addressed from psycho-cognitive and social perspectives. In this context, we argue for the relevance of integrating social-aware motion control into robotic systems. First, we identify the motion parameters that influence human perception and operator experience. Then, we implement a Model Predictive Control (MPC) framework that generates four distinct socially-informed robot behaviors. Finally, we conduct a user study to evaluate and validate these behaviors and assess their social impact on non-expert participants. The results demonstrate that variations in robot behavior significantly affect the perceived social acceptability of the system. These findings highlight the importance of incorporating human-centered considerations into motion generation strategies for robots operating in shared environments.

The rapid advancement of generative models has blurred the boundary between synthetic and real imagery, creating an urgent need for reliable deepfake detection. Yet most existing approaches rely on massive real--fake datasets, which are increasingly difficult to maintain as new generators continue to emerge. In this work, we investigate how much information about image authenticity is already encoded in modern multimodal vision representations. We find that frozen multimodal encoders naturally separate real and synthetic images in their embedding space, enabling a simple linear classifier to achieve strong performance without task-specific fine-tuning. Motivated by this observation, we develop a representation-aware data curation strategy that selects a compact set of representative generators for training. The resulting training set contains only 10K images, compared to 288K in AIGIBench and 4M in OpenFake, while improving robustness to unseen generators and distribution shifts. We additionally introduce RealWorldBench, a benchmark consisting of modern camera photographs, contemporary stock images, and outputs from recent commercial generators. Experiments across multiple benchmarks show that combining frozen multimodal representations with carefully curated training data provides a simple and effective approach to AI-generated image detection.

Large language models (LLMs) are increasingly evolving from simple text-based interaction systems into LLM agents that can maintain memory, use tools, access external environments, and execute tasks. As their capabilities and autonomy expand, the safety risks they face also become more diverse. Existing evaluations often rely on manually written scenarios, static prompts, or final-output judgments, making it difficult to capture the diverse risks that agents may face during task execution. We introduce ForesightSafety-SAGE, a fully automated scenario generation and safety evaluation framework for LLM agents. Based on five risk dimensions,we instantiae abstract and diverse safety risks in real-world task execution into 1,072 measurable evaluation scenarios. Using the automated evaluation pipeline, 12 LLM agents are evaluated under two authority contexts. The results show that current agents still face substantial behavioral safety risks during task execution, with an average ASR of 47.1% and several models exceeding 70%. These findings demonstrate the importance of executable, process-level evaluation for understanding and improving LLM agent safety.

Safety-aligned large language models often exhibit sycophancy, which is the tendency to affirm users' opinions regardless of factual accuracy. Although well-studied in English, its manifestation in other languages remains largely unexamined, leaving billions of non-English speakers potentially vulnerable to model-validated misinformation. We present the first large-scale, multi-model evaluation of cross-lingual sycophancy, benchmarking \textbf{six instruction-tuned models} across \textbf{1.1 million instances} spanning \textbf{38 languages} and \textbf{33 topic categories}. We identify a consistent resource-tier effect: sycophancy rates spike sharply in low-resource and zero-shot language settings. Critically, this degradation is topic-agnostic, as models fail uniformly across both benign and safety-critical prompts, offering no additional protection where it is most needed. We further identify tokenizer fertility as a structural driver of this alignment collapse. Collectively, our results demonstrate that prevailing alignment methodologies generalize poorly beyond high-resource languages, underscoring the urgent need for equitable multilingual safety techniques.

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%.

LLM-based scientific agents have shown strong capacity for autonomous research, yet their safety layers remain structurally divorced from core reasoning: they inspect pipeline outputs rather than shaping the deliberation that produces them. This separation opens two failure modes: safety signals accumulated at one stage are discarded before the next, and sequences of individually benign tool calls can compose into harmful outcomes that no single-step filter detects. To address these challenges, we introduce \textbf{SciTrace}, a framework that weaves safety reasoning into every stage of the scientific agent pipeline. SciTrace couples two complementary mechanisms: a \textit{Safety-Intrinsic Reasoning Loop} (SIR) that maintains a cumulative risk state across the Thinker, Experimenter, Writer, and Reviewer stages through joint task-and-safety deliberation, and a \textit{Compositional Tool-Chain Verifier} (CTV) that performs trajectory-aware safety checks before execution, catching risks that surface only across multi-step tool sequences. Evaluated on 240 high-risk research tasks and 120 tool-related risk tasks spanning six scientific domains, SciTrace achieves state-of-the-art (\textbf{SOTA}) safety among compared frameworks across four backbone models: it consistently improves tool call safety and adversarial robustness while preserving scientific output quality, and it uncovers \textbf{78.8%} of the compositional tool-chain escapes that single-step monitors miss. The project website is available at https://opensciagent.github.io/SciTrace/.

Large Language Model (LLM) safety has often been evaluated at the behavior level, which provides limited evidence of internal robustness, as these evaluations target outputs rather than representation-level vulnerability under intervention. We formalize this discrepancy as the audit gap: the difference between behavioral safety and robustness under intervention. To study this gap, we construct dissociated models that preserve safe outward behavior while remaining vulnerable in the latent space. We introduce an intervention-based evaluation framework to test model robustness through soft interventions in parameter and latent spaces, including harmful fine-tuning and layer-wise latent perturbations. To formalize the evaluation, we propose the Latent Vulnerability Score (LVS) to measure how easily harmful behavior can be elicited by bounded latent perturbations. Using this evaluation framework, we show that behavioral safety metrics are insufficient measures of representation-level robustness across multiple safely and unsafely aligned state-of-the-art models. Notably, dissociated models show substantially elevated LVSs despite comparable refusal behavior under harmful intervention, with intermediate representations being the most sensitive to intervention. Our results suggest that behavioral safety evaluation alone provides an incomplete picture of model robustness, motivating representation-aware audits of latent vulnerability and observable behavior.

Recent advances in generative AI, especially powerful Large Language Models (LLMs) and Large Reasoning Models (LRMs), raise concerns over the interpretability, safety and sustainability of these large and opaque AI models. The power of such architectures is derived not only from the scalability of deep neural networks, but also massively parallel hardware such as GPU clusters. The diffuse nature of deep neural networks gives them great function-approximation capability when provided with sufficient training data but imposes a cost in interpretability and computational efficiency. Observing that localised machine learning (ML) models tend to be more interpretable and computationally efficient than deep neural networks on small datasets, we reason by analogy that similar advantages may apply to specific localised hardware ML architectures. We argue that localised architectures with lower bandwidth but higher expressivity per node have the potential to be fundamentally more interpretable than deep neural networks running on GPU clusters while remaining competitive for smaller datasets. We then evaluate the suitability of various hardware ML paradigms for implementing such localised architectures and evaluate their per-node expressivity, energy efficiency and practical maturity of the technology required.

LLMs-as-judges are the only way to evaluate safety at scale. Despite their importance, LLM-judges themselves are rarely evaluated beyond human agreement in simple, static benchmarks. We therefore investigate two under-explored but crucial properties of LLMs-as-judges: their susceptibility to relying on in context-information, and their steerability to differing safety definitions, which may not align with their internal safety priors. We evaluate the safety judging abilities of many generalist LLMs and safety-specific judges, and investigate the impact of task demonstrations, novel in-context information, and changing safety definitions. We find that while LLM-judges can learn from new information, they are broadly unlikely to adjust their evaluations if the context or safety definition contradicts their prior.

Are tool-calling LLM agents equally safe throughout a conversation? We discover they are not: agents are most vulnerable at the very start of a session and become substantially safer after a few regular agentic tasks -- a phenomenon we term the cold-start safety gap. To study this systematically, we introduce Safety Over Depth for Agents (SODA), a benchmark that controls how many regular agentic tasks the agent completes before encountering a safety threat, supporting up to 20 preceding tasks. Evaluating 7 models from 4 families, safety improves by 9--52% as the number of preceding regular agentic tasks increases from zero to twenty. Representation analysis confirms that model hidden states gradually shift toward a safety-aligned region as more preceding tasks are present. By systematically studying which part of the preceding conversation matters most, we find that the regular agentic tasks themselves are the primary driver of safety, while the agent's own prior responses have less effect on safety but are essential for preserving later utility. This conclusion is further supported by evaluation on open-source safety benchmarks (AgentHarm, Agent Safety Bench) and utility benchmarks (BFCL, API-Bank), confirming that warming up the agent with regular agentic tasks before deployment makes it safer and preserves full capability. Based on these findings, we recommend a simple deployment strategy: having the agent complete a few regular agentic tasks before possible exposure to safety-critical requests mitigates the cold-start safety gap. Our code is available at https://github.com/Trustworthy-ML-Lab/Agent-Cold-Start-Safety-Gap

Reinforcement learning (RL) policies enable dynamic legged locomotion but lack mechanisms to avoid violations of safety constraints that are absent during training. Large-scale offline safe learning is impractical for covering all edge cases. Existing safety frameworks either rely on reduced-order models that cannot reason about whole-body behaviors or require conservative recovery controllers that degrade task performance. We propose a predictive safety filter that post-hoc filters the nominal contact locations fed to the RL policy. When a collision is predicted, a sampling-based optimizer asynchronously searches for safer contact sequences using a full-physics model, while a learned value function bootstraps long-horizon returns. Our three algorithmic components (geometric projection of sampled contacts, momentum-augmented updates, and replica-exchange) make the optimization tractable in a discontinuous contact landscape. We validate the filter on a quadruped robot in dense, cluttered environments, both in simulation and in the real world, showing substantial reductions in safety violations with minimal deviation from the nominal input.

Safety is a fundamental requirement in the development of autonomous driving (AD) systems. While function offloading has demonstrated significant benefits in terms of computational efficiency and energy consumption, its application to safety-critical AD functionality introduces new challenges. In particular, offloaded service compositions incur increased and variable response times due to wireless vehicle-to-everything (V2X) communication, which directly affects the vehicle's reaction time and thus its safety guarantees. In this paper, we address this challenge by extending the definitions of Responsibility-Sensitive Safety (RSS) to explicitly account for different response times of local and offloaded AD service compositions. Based on this extension, we propose an integration into function offloading, using the RSS safety constraints for offloading decision-making and fallback mechanisms. Offloaded service compositions are only permitted if the current traffic situation remains safe under the corresponding end-to-end response time. If this condition is violated, the system performs a controlled fallback to local execution. Furthermore, we introduce an enhanced fallback strategy that includes a warm-standby phase for offloaded services, enabling faster and safer transitions from offloaded to local services. The proposed approach is integrated into our AD stack and evaluated in both simulation and the real world. Experimental results demonstrate that the proposed method improves safety compared to state-of-the-art function offloading and safety frameworks, while preserving the benefits of distributed computation when safety conditions allow.

Safety alignment for large language models relies on preference data, but current pipelines often train on large, redundant datasets. Existing data selection methods typically score each preference pair independently, collapsing directional preference information into scalar quality or diversity scores. This sample-centric view is especially limiting in multi-dataset settings, where shared safety directions coexist with dataset-specific residual risks. We propose DOG-DPO, a training-free data selection framework that treats preference pairs as structured geometric signals. DOG-DPO first represents each preference pair as a direction in model representation space. It then decomposes multi-dataset preference geometry into a global anchor subspace and dataset-specific residual subspaces. Finally, it selects subsets by maximizing diversity-based coverage, encouraging broad, non-redundant coverage of alignment directions before DPO training. Across six safety benchmarks and two model backbones, DOG-DPO achieves a strong utility-robustness trade-off using only 11% of the preference pairs. It recovers most of the safety gains of full-data training while remaining entirely teacher-free, training-free, and substantially faster than representative selection baselines.

Large audio language models (LALMs) are increasingly deployed in real-world applications, yet their safety alignment is still primarily evaluated on monolingual, text-based harmful prompts. This leaves their generalizability under multilingual and spoken settings, particularly code-switched speech, largely underexplored. To address this gap, we introduce SpeechJBB, an audio jailbreak dataset for benchmarking across multiple state-of-the-art LALMs. The extent of safety weaknesses is further probed by introducing an augmented setting where phonologically plausible pseudo-words are inserted around safety-critical terms to simulate localized obfuscation. Across models, code-switched harmful audio yields substantially high jailbreak success rates (JSR), with non-English monolingual and non-English code-switched pairs exhibiting the highest attack success. Pseudo-word insertion further reduces refusal rates, which demonstrates that natural-sounding obfuscation can effectively bypass safety policies.

Consistency training encourages models to behave similarly across different contexts, and has shown promise for reducing misalignment. We broaden the scope of consistency training in two ways. First, we introduce two new internal consistency targets: MLP Consistency Training (MLPCT), which matches post-activation MLP states, and Attention Consistency Training (AttCT), which matches per-head attention distributions. Second, we apply consistency training to four additional safety threats: persona in-context learning attacks, adversarial frustration, prefill attacks, and conditional misalignment. Across several models and threat settings, we find that consistency training reduces misalignment well beyond the sycophancy and jailbreak settings studied in prior work. We also find cases of cross-threat generalization, where training against one failure mode improves robustness to another, and identify a shared residual-stream mechanism underlying ACT, MLPCT, and AttCT, while distinguishing BCT as mechanistically distinct. Our results suggest that consistency training is a flexible and extensible framework for alignment, capable of unifying defenses against a broader class of model pathologies.

Embodied AI systems are increasingly expected to reason and act over extended horizons in physical environments. This growing capability brings safety to the foreground, because failures in the physical world can harm people, damage objects, and disrupt workplaces. Although safe embodied AI has attracted substantial attention, the literature remains fragmented across planning, policy design, and runtime execution. Long-horizon robotic manipulation is a particularly revealing anchor domain for this problem because semantic misgrounding, subtask-level error propagation, execution drift, and contact-rich physical risk can accumulate within the same closed-loop system. This survey therefore provides a structured review of safety in long-horizon robotic manipulation from an embodied AI perspective. We organize the literature by intervention locus, covering planning-time, policy-time, and execution-time safety, and we analyze the strength of the evidence that each line of work provides, distinguishing formal guarantees, statistical support, and empirical safety heuristics. This framework clarifies the distinct roles of backbone capability papers, direct safety mechanisms, and benchmark or evaluation studies, while exposing where current safety claims are well supported and where they remain indirect. We identify persistent gaps, including limited evidence for policy-time safety, weak formal support for contact-rich long-horizon manipulation, immature uncertainty-triggered intervention, and a shortage of manipulation-specific safety benchmarks. We conclude by outlining research directions for cross-layer assurance, evaluation design, and safer deployment of long-horizon robotic agents in real-world settings.

Large language models (LLMs) are rigorously aligned to refuse harmful requests, a process that inherently cultivates a latent capacity to evaluate and recognize unsafe content. In this work, we reveal that this advanced safety awareness inadvertently introduces a fatal vulnerability. We introduce Posterior Attack, a single-query jailbreak that bypasses guardrails by prompting the model to generate the exact harmful response its internal classifier would normally flag as unsafe. Through extensive empirical evaluation across 30 open-source LLMs (up to 35B parameters in size) and frontier models (e.g., GPT-5, Claude 4.6), we observe a striking phenomenon: models with superior safety-judgment capabilities are disproportionately more susceptible to this exploitation. To explain this, we formalize the Safety Paradox, analytically showing that monotonic improvements in safety alignment naturally amplify posterior vulnerability. Finally, we establish a causal link via reinforcement learning interventions, exemplifying that artificially degrading a model's safety judgment immunizes it against the attack, whereas enhancing judgment exacerbates the vulnerability. Our findings highlight potential flaws in current alignment paradigms, indicating that defense mechanisms may require further structural refinement.