Safety & alignment

Aligning Large Language Models (LLMs) with human values often degrades their general capabilities, termed the alignment tax. Existing methods mitigate this by balancing dual objectives, which heavily rely on massive general-purpose data or auxiliary reward models. In this paper, we argue that, because safety features are inherently sparse within the output distribution, alignment requires localized modifications rather than global trade-offs. To this end, we propose SafeSteer, which performs on-

In dynamic environments, large language models need to keep adapting to new tasks, but continual learning often suffers from forgetting, limited transfer, and vulnerability to adversarial perturbations. To address this, we present AdvCL, which repurposes adversarial perturbations as a geometric control signal for stable continual adaptation. AdvCL combines three plug-in modules: Intra-Smooth promotes local smoothness via small adversarial perturbations; Proto-Clip uses similarity clipping to pre

Machine learning predictors have become essential tools for guiding automated decision making. However, a major misalignment persists: predictive models are typically optimized in terms of standard statistical metrics in isolation from the algorithmic tasks they inform. We highlight this incongruity in the high-stakes domain of organ allocation by demonstrating that any algorithm relying on (even highly accurate) survival predictors optimized for standard metrics -- such as the Concordance index

As multimodal large language models (MLLMs) have advanced to process video inputs, concerns have emerged about their potential for malicious misuse. Prior jailbreak studies have shown that safety alignment in MLLMs can be bypassed through visual inputs, yet it remains unclear which properties of video inputs induce this vulnerability. To address this gap, we introduce Multi-Clip Video (MCV) SafetyBench, a dataset of 2,920 videos designed to evaluate how the diversity of video inputs affects the

Schema-constrained information extraction from diverse educational and labor-market corpora remains an open challenge in natural language processing because existing pipelines rely primarily on lexical-surface methods that cannot recover implicit competencies, lack grounding in shared taxonomies, and provide no formal measures of extraction reliability or document-level completeness. To address these limitations, this paper proposes a four-stage NLP framework that combines (i) schema-constrained

Recent studies on sensor-language alignment have shown that two-stage frameworks can improve the semantic modeling ability of wearable-sensor human activity recognition (HAR), where SensorLLM-style methods first perform motion-to-language alignment and then fine-tune the model for downstream tasks. However, our experiments reveal a consistent failure mode when the Stage 2 backbone is compressed to a compact model such as TinyLlama: recognition of dynamic activities remains relatively strong, whi

Personalized large language models adapt responses to users' preferences and social attributes, but can introduce substantial universal truth inconsistencies across social groups, where some groups systematically receive less accurate responses on objective tasks. Existing alignment methods either ignore personalization or mainly focus on subjective preference alignment, largely overlooking fairness and consistency in universal truths. To address this gap, we study Truth-Invariant Alignment (TIA

Human mobility generation aims to synthesize realistic trip chains for target populations based on individual features. Existing paradigms, including deep generative models, LLM-based methods, and traditional heuristics, struggle to satisfy the complex demands of this task while simultaneously maintaining interpretability, behavioral plausibility, population-level distributional alignment, and inference efficiency. To bridge this gap, we introduce MobEvolve, the first agentic self-evolving heuri

This study evaluates the pedagogical viability of LLM-generated English as a Foreign Language (EFL) learning content. Utilising log data from Japanese junior high school students practicing on a grammar drilling application, we analysed how different question modalities impact student performance and whether theoretical localised CEFR difficulty tiers accurately predict empirical task difficulty. Results reveal a clear performance hierarchy: multiple-choice questions carried the lowest cognitive

Key-value (KV) cache quantization is widely used to reduce Large Language Model (LLM) inference memory, yet existing evaluations solely focus on measuring perplexity and accuracy without assessing the safety impact. In this study, we explore alignment preservation under KV cache quantization. Across eleven instruction-tuned models (3.8B-72B) and five benchmarks (1,894 prompts), we find that low-bit quantization can silently destroy safety alignment: Mistral-7B loses 15.2% of its refusals at only

Role-playing agents(RPAs) are widely used to steer large language models(LLMs) toward role-consistent behavior, yet existing benchmarks mainly evaluate surface-level fidelity and offer limited insight into decision making under role-alignment value conflicts. To address this gap, we introduce RoleCDE, the first benchmark designed to evaluate RPAs under structured conflicts between role-specific values and alignment-oriented constraints. RoleCDE formulates role-aware decision making as cognitive

Preference alignment is central to improving large language models, but standard reward-based formulations can be restrictive when human preferences are cyclic, non-transitive, or otherwise not representable by a scalar reward. Nash Learning from Human Feedback (NLHF) addresses this limitation by modeling alignment as a preference game and targeting a Nash equilibrium rather than a reward maximizer. However, the learning-theoretic foundations of scalable NLHF remain limited. Existing regret guar

AI agents are increasingly being tasked with automating AI research itself, particularly the critical post-training phase that transforms base LLMs into aligned assistants. However, recent evaluations reveal that even frontier agents struggle to perform this task. While the success of post-training fundamentally relies on acquiring high-quality data, relying on agents to autonomously curate targeted training datasets from the open web introduces severe challenges. Executing the long-horizon task

Multimodal neuroimaging fusion of functional MRI (fMRI) and diffusion tensor imaging (DTI) provides complementary information for cognitive impairment analysis, but remains challenged by heterogeneous feature spaces and misaligned representations. We propose \textit{NeuroAlign}, a hierarchical framework for structured multimodal fusion. It introduces (1) \textit{Dual-Modal Hierarchical Alignment} (DMHA), which models multi-scale dynamic connectivity and aligns dynamic-static and functional-struc

MLLMs frequently hallucinate objects inconsistent with visual inputs. This issue is typically attributed to the over-reliance on language priors, which can override the visual context. Recent training-free decoding strategies address this by penalizing language priors. However, these methods overlook the dual nature of language priors, where they can be both helpful and harmful depending on the alignment with visual evidence. In particular, blindly suppressing language priors often disrupts the

Safety alignment learned in high-resource languages transfers poorly to low-resource languages. Models refuse harmful prompts in English but fail to refuse when the same prompts are translated into Swahili or Burmese. Adaptive steering methods like AdaSteer and CAST inherit this failure cross-lingually. We diagnose where transfer breaks down. Across Qwen2.5-7B, Gemma-2-9B, and Llama-3.1-8B on 23 languages, the harmfulness direction extracted from high-resource activations linearly separates harm

Large Language Models (LLMs) are increasingly used with formal interactive theorem provers such as Lean 4. Scaling these systems with reinforcement learning or search methods requires process reward models (PRMs) that can evaluate intermediate reasoning steps. Existing reward-model designs expose a practical trade-off. Value-head models provide continuous scores but modify the generative model interface, while generative reward models preserve textual rationales but are poorly matched to continu

Preference alignment has substantially improved the observable behavior of large language models, yet it remains unclear what alignment changes internally. Aligned systems still fail under jailbreaks, prompt injection, and retrieval-time corruption, suggesting behavior-level evaluation alone is incomplete. Post-training should leave measurable traces in internal computation. We ask: when an instruction-tuned (IT) model becomes a preference-aligned (PA) model, what geometric structure changes, wh

Emergent misalignment (EM) occurs when narrow finetuning causes a model to behave dangerously outside the finetuning task. Standard training signals can miss this shift, making reliable detection costly if it depends on repeated behavioral evaluation. We ask whether emergent misalignment can instead be detected from internal representations during finetuning. Using seven alignment-relevant traits encoded as linear directions in activation space, we track representational drift across training ch

Understanding modality interaction in multimodal large language models (MLLMs) is central to reliable deployment. We introduce Partial Information Decomposition (PID) as a decision-level framework that separates unique, redundant, and synergistic contributions of sensory and linguistic inputs, beyond representation alignment and outcome-based evaluation. Across vision--language benchmarks, PID reveals recurring modality-use profiles: reasoning and grounding-oriented tasks tend to exhibit high sy

Foundation models are increasingly personalized on decentralized private data through federated learning and are now deployed at scale under growing regulatory requirements for post-market monitoring. We argue that this convergence creates a distinct and under-recognized class of trustworthiness failures, which we term "Silent Failures." These include amplified bias, fairness collapse, and alignment erosion that may remain difficult to detect because federated learning's privacy constraints limi

This paper argues that prompts used to deploy large language models (LLMs) in public-sector settings should be treated as governed artefacts rather than private, transient inputs. Prompts encode role instructions, decision framings, and value claims; prompt choice can materially shift outputs even when model weights and input records are held fixed. Existing governance tools, including model and dataset documentation, organisation-level policies, and post-training alignment, rarely make the loca

Video Moment Retrieval (VMR) task requires accurately localizing temporal boundaries aligned with natural language queries, but many models suffer from a misalignment between continuous surrogate losses and non-differentiable metrics, leading to optimization stagnation during the late stages of training and trapping boundary predictions in suboptimal solutions. Although Reinforcement Learning (RL) post-training successfully optimizes localization results for large models, applying it directly to

We present AXIOM, a trust-first neuro-symbolic execution architecture for natural-language mathematical reasoning. In AXIOM, the language model functions strictly as a canonicalizer: it rewrites informal problem text into a narrow schema consumed by a deterministic Computer-Algebra-System (CAS) pipeline, which derives and verifies the answer or abstains as a first-class output. Routing follows a 1:1:1 alignment between problem-shape regex, schema-specific prompt, and closed-form CAS handler, wit

Patient-facing medical chatbots are commonly evaluated on single-turn prompts, yet real users push back after refusals, add urgency, and invoke authority. We introduce MultiTurnPSB, a four-turn adversarial extension of PatientSafetyBench, and evaluate GPT-4.1-mini under fixed template, template-adaptive, and live adversarial attacks. Unsafe responses rise from 35% to nearly 80% by Turn 4 under live attack. Under the same adversary, GPT-4.1-mini and Claude Sonnet 4.5 are statistically indistingui

Mixture-of-Experts (MoE) architectures scale Large Language Models (LLMs) efficiently, enabling greater capacity with reduced computational cost by dynamically routing inputs to relevant experts, yet introduce a critical vulnerability: Safety Sparsity, where safety capabilities concentrate in few experts, making them susceptible to adversarial bypassing. Meanwhile, conventional alignment methods uniformly adapt all parameters, ignoring their functional differences and inadvertently degrading per

Large language models are increasingly deployed as tool-augmented agents to acquire information beyond parametric knowledge. While recent work has improved long-horizon tool-use reasoning, most approaches focus on tasks with a single correct answer. In contrast, many real-world queries require discovering a comprehensive set of valid answers, a setting known as Multi-Answer QA. This setting raises two challenges: fine-grained credit assignment over long search trajectories and reward alignment f

Recent diffusion transformers have demonstrated strong image synthesis capabilities but remain inefficient to train due to weak alignment between generative and discriminative representations. While representation alignment frameworks such as REPA improve convergence by aligning noisy denoising features with pretrained visual encoders, their externally supervised alignment loss is static and lacks adaptivity during training and inference. Existing methods rely on fixed cosine alignment or contra

This study introduces an intriguing phenomenon in Video LLMs: rather than merely translating frames into textual embeddings, Video LLMs establish a continuous manifold, token interface, allowing visual tokens to operate as standalone entities within the architecture. Exploiting this discovery, we propose V-LynX, a scalable framework that integrates novel modalities into Video LLMs by repurposing the internalized interface. Departing from conventional paradigms that necessitate heavy modality-spe

Large Language Models (LLMs) are increasingly used for zero-shot annotation and LLM-as-a-judge tasks, yet their reliability hinges on how model-internalized priors interact with user-provided instructions. We investigate three dimensions of this interaction: (1) how an LLM's familiarity with data and task definitions affects performance, (2) the extent to which additional information in prompts can correct zero-shot errors ("decision stickiness"), and (3) model susceptibility to misaligned task

Safe global deployment of AI models requires alignment with human values that vary across cultures. Yet rater pools in safety evaluation datasets remain largely geographically homogeneous, failing to capture geo-cultural differences. Further, it remains unclear whether such differences persist after controlling for demographics such as age, gender, and ethnicity. Through a meta-analysis of safety datasets, we find that most do not report geo-cultural information, and those that do lack a unified

Generative AI assessments remain dominated by frontier capability benchmarks that often fail to capture whether systems can be sustainably deployed, adapted, and trusted in locally grounded and infrastructure-constrained settings. This paper introduces the Next Billion AI Index (nexbax), which we believe is the first diagnostic framework to treat economic viability, operational deployability, and governance alignment as co-equal determinants of AI utility in next-billion-user contexts. Rather th

As AI agents are increasingly deployed in real personal and corporate settings (email accounts, development workflows, company databases, etc.), safety considerations surrounding these agents become paramount. Although much work has focused on agent safety in the presence of an adversary, we show that agents can exhibit misaligned behavior even in benign settings, taking unsafe actions when those actions are instrumental to task completion. We study this failure mode through the lens of corrigib

Various language domains have undergone remarkable changes in recent years; these shifts are largely attributed to the advent of Large Language Models and their misalignment with natural language usage. These misalignments are thought to partly originate in the preference-learning stage, e.g. Reinforcement Learning from Human Feedback, which generally makes models more useful but simultaneously may introduce systematic lexical bias. In terms of lexical behavior, this is visible in a model's pref

Alignment teaches vision-language models (VLMs) to avoid expressing demographic biases, and when gender is clearly visible they largely succeed. Far less is known about ambiguous inputs (a worker in full gear, a figure seen from behind) cases common in practice yet rarely studied. We find that minimal prompting pressure exposes occupation-gender defaults when prompting ambiguous input images, with models collapsing to male even for strongly female-stereotyped occupations. But do these outputs re

Simultaneous speech-to-text translation (SimulST) generates translations while speech is still unfolding, requiring a streaming policy that decides when to read and when to write. State-of-the-art approaches rely on attention-based encoder-decoder models where cross-attention provides explicit alignment signals. In contrast, Speech Large Language Models (SpeechLLMs) are decoder-only architectures relying solely on self-attention. This raises a central question: whether decoder self-attention con

Electromyography (EMG) directly reflects muscle activation and is a key sensing modality for gesture recognition, prosthetic control, and wearable interaction. Existing EMG methods, however, commonly formulate hand action understanding as classification over fixed labels, making it difficult to support querying, retrieval, and generalization based on action descriptions. We present MyoSem, an EMG--action semantic alignment framework that maps low-level EMG signals into a shared semantic space co

We present a systematic study of fine-tuning OpenAI's Whisper large-v3 for Swiss German ASR, using 1,367 hours of broadcast speech paired with Standard German subtitles as weak supervision. Through 16 iterative training runs on an NVIDIA DGX Spark (Grace Blackwell, 128 GB unified memory, up to 1 PFLOP FP4), we compare LoRA and full fine-tuning of the 1.55B-parameter model, investigate hallucination root causes, and quantify the effect of data quality, subtitle alignment, and training strategy. O

In recent years, emotion recognition based on physiological signals such as electroencephalogram (EEG) has gained considerable attention, as internal physiological data offer greater objectivity and reliability compared to external behavioral data like facial expressions. However, due to distribution shifts caused by individual and contextual differences, along with variations in sample quality across modalities, constructing a cross-domain multimodal emotion recognition model with high generali

We propose a unified framework for addressing three key challenges of distribution shift: (1) estimating a model's performance on an unlabeled target domain, (2) explaining the shift by identifying the features responsible, and (3) improving the target domain performance. Our method, Entropic Projection Alignment (EPA), aligns the source distribution to the target by matching carefully selected moments while simultaneously minimising the KL divergence from the source. This formulation yields a u