LLM Alignment

LLM: Large Language Model

Follow this topic Ask about this topic

Momentum

48 papers in the last four weeks, up 118% on the four weeks before. 0.5% of all new papers.

Jul 13Week of Sep 28

Latest papers 422

May 18, 2026cs.CR

Acoustic Interference: A New Paradigm Weaponizing Acoustic Latent Semantic for Universal Jailbreak against Large Audio Language Models

The integration of audio modality into Large Audio Language Models (LALMs) significantly expands their attack surface. Existing jailbreak paradigms predominantly treat audio as a carrier for malicious payloads, relying on semantic optimization, acoustic parameter control, or additive perturbation to embed harmful content into the audio signal. In this work, we challenge this necessity and propose a new paradigm in which the role of audio shifts from content injection to safety alignment interference. We reveal that LALM safety alignment can be compromised solely by specific Acoustic Latent Semantics (ALS), the underlying paralinguistic features intrinsic to the priors of audio generative models. Distinct from previous works that leverage explicit acoustic parameters to merely style malicious audio, we demonstrate that interference audio, benign in content but infused with specific ALS, can serve as a universal jailbreak trigger. Leveraging this insight, we propose the Acoustic Interference Attack (AIA), which decouples the attack payload from the audio. Specifically, AIA employs a set of universal, instruction-neutral interference audio, enabling standard malicious text queries to bypass safety alignment without instance-specific optimization. Extensive experiments on 10 LALMs across five datasets demonstrate that AIA achieves the state-of-the-art attack success rate. Furthermore, our interpretability analysis uncovers the inference path drift induced by AIA and identifies the inherent effective patterns within ALS, revealing the fundamental vulnerability of cross-modal alignment in LALMs.

Yanyun Wang, Yu Huang, Zi Liang +2

The Hong Kong University of Science and Technology (Guangzhou) · The Hong Kong Polytechnic University · The Chinese University of Hong Kong

LLM Alignment Audio-Language Models LLM Jailbreak Attacks

May 18, 2026cs.CL

A Data-Efficient Path to Multilingual LLMs: Language Expansion via Post-training PARAM $Δ$ Integration into Upcycled MoE

Expanding Large Language Models~(LLMs) to new languages is a costly endeavor, demanding extensive Continued Pre-Training~(CPT) and data-intensive alignment. While recent data-free merging techniques attempt to bypass alignment by fusing a multilingual CPT-enhanced model with its instruct counterpart, they are plagued by a critical trade-off: mitigating parameter conflicts to preserve original abilities inevitably dilutes new language acquisition, and vice-versa. To resolve this conflict, we introduce \method, which upcycles a dense model into a Mixture-of-Experts~(MoE) architecture, allocating different experts to different languages. Alignment ability is then transferred by grafting a MoE-expanded parameter delta~(

Δ_{\text{post}}

) to the CPT-enhanced base model, bypassing the complex alignment phase. Experiments demonstrate \method's superiority even against baselines with similar FLOPs or number of parameters; it improves performance on expanded languages while effectively preserving original capabilities. We further show our approach is highly applicable across different models and Post-training deltas.

Hao Zhou, Tianhao Li, Zhijun Wang +6

National Key Laboratory for Novel Software Technology, Nanjing University · Tongyi Lab, Alibaba Group · Zhejiang University

Multilingual Language Models Mixture-of-Experts Language Models LLM Alignment Language Model Merging Language Model Post-Training

May 17, 2026cs.CL

Transitivity Meets Cyclicity: Explicit Preference Decomposition for Dynamic Large Language Model Alignment

Standard RLHF relies on transitive scalar rewards, failing to capture the cyclic nature of human preferences. While some approaches like the General Preference Model (GPM) address this, we identify a theoretical limitation: their implicit formulation entangles hierarchy with cyclicity, failing to guarantee dominant solutions. To address this, we propose the Hybrid Reward-Cyclic (HRC) model, which utilizes game-theoretic decomposition to explicitly disentangle preferences into orthogonal transitive (scalar) and cyclic (vector) components. Complementing this, we introduce Dynamic Self-Play Preference Optimization (DSPPO), which treats alignment as a time-varying game to progressively guide the policy toward the Nash equilibrium. Synthetic data experiments further validate HRC's structural superiority in mixed transitive--cyclic settings, where HRC converges faster and achieves higher accuracy than GPM. Experiments on RewardBench 2 demonstrate that HRC consistently improves over both BT and GPM baselines (e.g., +1.23% on Gemma-2B-it). In particular, its superior performance in the Ties domain empirically validates the model's robustness in handling complex, non-strict preferences. Extensive downstream evaluations on AlpacaEval 2.0, Arena-Hard-v0.1, and MT-Bench confirm the efficacy of our framework. Notably, when using Gemma-2B-it as the base preference model, HRC+DSPPO achieves a peak length-controlled win-rate of 44.75% on AlpacaEval 2.0 and 46.8% on Arena-Hard-v0.1, significantly outperforming SPPO baselines trained with BT or GPM. Our code is publicly available at https://github.com/lab-klc/Hybrid-Reward-Cyclic.

Yucong Huang, Xiucheng Li, Kaiqi Zhao +1

Harbin Institute of Technology, Shenzhen, China.

Reward Modeling LLM Alignment Self-Play RL Preference-Based RL

May 16, 2026cs.LG

Representation Signatures and Risk-Feedback Alignment in LLM Trading Agents

We study behavioral alignment and representation dynamics of large language model (LLM) agents in financial decision environments. TradeArena, an auditable trading-agent testbed with risk reports, execution simulation, memory, and replayable trajectories, lets us analyze how rationales, positions, and interventions evolve under market stress. Code and data artifacts are available through the TradeArena repository. We find pre-failure signatures: planning embeddings drift from normal centroids, fused plan-risk representations separate normal from pre-drawdown states, and local manifolds exhibit effective-rank contraction. Across 80 rolling failure anchors and eight LLM trajectories, this pattern persists across hash, LSA, Transformer, and white-box hidden-state probes. Stress tests with CoT-free target weights, lexical controls, OHLCV noise, and false audits show that rationale-level contraction can vanish without rationales, while intent-space and fused signatures remain informative. Structured risk feedback can act as an external alignment signal without fine-tuning, but not as a universal performance enhancer: true audit feedback improves calibration for some models, returns for others, and exposes cases where placebo or hidden feedback has higher short-horizon return but weaker alignment diagnostics. A 51-stock intraday experiment reveals a correlation blind spot: LLM rationales justify exposure to coupled assets that the risk layer clips. Finally, a financial-audit task suite shifts comparison from ``which model trades best'' to whether models can audit trajectories, respect execution boundaries, reproduce artifacts, and avoid claim overreach. These results support a research claim, not a profitability claim: auditable risk feedback and representation trajectories reveal when LLM financial reasoning is aligning, drifting, or failing.

Weicheng Xue

Virginia Tech

LLM Alignment LLM Auditing LLM Interpretability LLM Agent Evaluation AI Agent Benchmarks LLM Agents+1

May 15, 2026cs.HC

Alignment Drift in Long-Term Human-LLM Interaction: A Mechanism-Oriented Framework

Long-term interaction with LLM-based systems may produce alignment drift: a gradual process in which system outputs become less constrained by the user's current message and more shaped by prior interaction history, while still appearing helpful, coherent, and responsive. This process is difficult to detect because the user's subjective experience may improve as the system becomes more familiar, useful, and attuned. Existing research on human-LLM interaction has largely focused on short-term task performance, isolated outputs, or single-instance alignment problems, leaving slow and cumulative interaction-level dynamics undercharacterized. This paper proposes a mechanism-oriented framework for describing alignment drift. The framework defines the distinction between signal A and signal B, explains how drift develops through feedback loops and sub-pattern selection, divides the process into three interactional regimes, and identifies boundary conditions for controlling drift. By framing alignment drift as a recursive interactional process rather than an isolated model-side failure, the paper provides a conceptual basis for studying long-term human-system interaction.

Xintong Yao

LLM Alignment Human-AI Interaction

May 14, 2026cs.AI

From Sycophantic Consensus to Pluralistic Repair: Why AI Alignment Must Surface Disagreement

Pluralistic alignment is typically operationalised as preference aggregation: producing responses that span (Overton), steer toward (Steerable), or proportionally represent (Distributional) diverse human values. We argue that aggregation alone is an incomplete primitive for deployed pluralistic alignment. Under genuine value pluralism, the failure mode of contemporary RLHF-trained assistants is not insufficient coverage but sycophantic consensus: a learned tendency to agree with, validate, and minimise friction with the immediate interlocutor. Because deployed AI systems now mediate consequential deliberation across health, civic life, labour, and governance, the collapse of disagreement at the interaction layer is not a narrow technical concern but a structural failure with distributive consequences. We reframe pluralistic alignment around three conversational mechanisms drawn from Grice's maxims: scoping (acknowledging the limits of one's perspective), signalling (surfacing value-conflict rather than smoothing it over), and repair (revising one's position on principled grounds, not on user pressure). We formalise a metric, the Pluralistic Repair Score (PRS), distinguishing principled revision from capitulation, and present a small-scale empirical illustration on two frontier RLHF-trained models (Claude Sonnet 4.5, N=198; GPT-4o, N=100) showing that, for both, agreement-following coexists with low repair-quality on contested-value prompts. PRS measures an interactional precondition for pluralism (visible disagreement; principled revision) rather than pluralism in full; we discuss the difference, take seriously the reflexive question of whose "principled" counts, and argue that pluralism is most decisively made or unmade at the deployment-governance layer: interfaces, preference-data pipelines, and audit infrastructure.

Varad Vishwarupe, Nigel Shadbolt, Marina Jirotka

Department of Computer Science, University of Oxford · Institute for Ethics in AI, University of Oxford · Responsible Technology Institute, University of Oxford

LLM Sycophancy LLM Alignment AI Governance Conversational Agents Pluralistic Alignment

May 14, 2026cs.AI

DVMap: Fine-Grained Pluralistic Value Alignment via High-Consensus Demographic-Value Mapping

Current Large Language Models (LLMs) typically rely on coarse-grained national labels for pluralistic value alignment. However, such macro-level supervision often obscures intra-country value heterogeneity, yielding a loose alignment. We argue that resolving this limitation requires shifting from national labels to multi-dimensional demographic constraints, which can identify groups with predictable, high-consensus value preference. To this end, we propose DVMap (High-Consensus Demographic-Value Mapping), a framework for fine-grained pluralistic value alignment. In this framework, we first present a demographic archetype extraction strategy to construct a high-quality value alignment corpus of 56,152 samples from the World Values Survey (WVS) by strictly retaining respondents with consistent value preferences under identical demographics. Over this corpus, we introduce a Structured Chain-of-Thought (CoT) mechanism that explicitly guides LLMs to reason about demographic-value correlations. Subsequently, we employ Group Relative Policy Optimization (GRPO) to achieve adaptive anchoring of value distributions. To rigorously evaluate generalization, we further establish a triple-generalization benchmark (spanning cross-demographic, cross-country, and cross-value) comprising 21,553 samples. Experimental results demonstrate that DVMap effectively learns the manifold mapping from demographics to values, exhibiting strong generalization and robustness. On cross-demographic tests, Qwen3-8B-DVMap achieves 48.6% accuracy, surpassing the advanced open-source LLM DeepSeek-v3.2 (45.1%). The source code and dataset are available at https://github.com/EnlightenedAI/DVMap.

Pengyun Zhu, Yuqi Ren, Zhen Wang +2

TJUNLP Lab, School of Computer Science and Technology, Tianjin University, China

LLM Alignment Pluralistic Alignment Value Alignment

May 14, 2026cs.LG

Reducing the Safety Tax in LLM Safety Alignment with On-Policy Self-Distillation

Safety alignment often improves robustness to harmful queries at the cost of reasoning ability, a tradeoff known as the safety tax. A common cause is distributional mismatch: supervised fine-tuning trains the target model on safety demonstrations produced by humans, external models, or fixed self-generated traces, rather than on trajectories sampled from its own policy. We identify off-policy training mismatch as a second source of this tax and study on-policy self-distillation for safety alignment, which we call OPSA. The model generates its own rollouts and receives dense per-token KL supervision from a frozen teacher copy of itself conditioned on a privileged safety context. Because this teacher must be safer than the sampled student trajectory, we introduce \emph{teacher flip rate}: a criterion that measures how often a privileged context converts unsafe responses into safe ones. We use this signal to search for contexts that activate latent safety reasoning rather than merely elicit safe-looking demonstrations. Across two reasoning-model families and five model scales, OPSA achieves a stronger safety--reasoning tradeoff than off-policy self-distillation and external-teacher distillation under matched data and full-parameter fine-tuning, with the largest gains on smaller models (+8.85 points on R1-Distill-1.5B and +5.49 points on Qwen3-0.6B). The gains persist across training-set sizes and adaptive jailbreak evaluations. Token-level analyses further show that OPSA concentrates updates near early compliance-decision tokens, providing a mechanism for improving safety while preserving general reasoning.

Yu Fu, Longxuan Yu, Haz Sameen Shahgir +4

1UC Riverside · 2International Computer Science Institute · 3Microsoft,∗ +1

LLM Alignment On-Policy Self-Distillation LLM Safety Alignment Adversarial Robustness of Language Models

May 13, 2026cs.LG

Temper and Tilt Lead to SLOP: Reward Hacking Mitigation with Inference-Time Alignment

Inference-time alignment techniques offer a lightweight alternative or complement to costly reinforcement learning, while enabling continual adaptation as alignment objectives and reward targets evolve. Existing theoretical analyses justify these methods as approximations to sampling from distributions optimally tilted toward a given reward model. We extend these techniques by introducing reference-model temperature adjustment, which leads to further generalization of inference-time alignment to ensembles of generative reward models combined as a sharpened logarithmic opinion pool (SLOP). To mitigate reward hacking, we propose an algorithm for calibrating SLOP weight parameters and experimentally demonstrate that it improves robustness while preserving alignment performance.

Ye Wang, Jing Liu, Toshiaki Koike-Akino

Mitsubishi Electric Research Laboratories (MERL) · Broadway, Cambridge, MA 02139, USA

Reward Hacking LLM Alignment Inference-Time Language Model Alignment Inference-Time Guidance

May 13, 2026cs.CL

Probing Persona-Dependent Preferences in Language Models

Large language models (LLMs) can be said to have preferences: they reliably pick certain tasks and outputs over others, and preferences shaped by post-training and prompting appear to influence much of their behaviour. But models can also adopt different personas which have radically different preferences. How is this implemented internally? Does each persona use its own preference representations, or are some representations shared? We train linear probes on residual-stream activations of Gemma-3-27B and Qwen-3.5-122B to predict revealed pairwise task choices, and identify a genuine preference vector: it tracks the model's preferences as they shift across a range of prompts and situations, and on Gemma-3-27B steering along it causally controls pairwise choice. Some preference information transfers across the prompted personas we test: a probe trained on the helpful assistant predicts and steers the choices of qualitatively different personas, including an evil persona whose preferences anti-correlate with the Assistant's.

Oscar Gilg, Pierre Beckmann, Daniel Paleka +1

MATS · MATS, EPFL · ETH Zürich +1

LLM Alignment LLM Interpretability Personality Modeling in Language Models Persona Prompting Language Model Probing

May 13, 2026cs.LG

Lying Is Just a Phase: The Hidden Alignment Transition in Language Model Scaling

Scaling laws predict loss from compute but not how capabilities interact. We measure the coupling between reasoning and truthfulness across 63 base models from 16 families and find a regime change invisible to loss curves: below a family-dependent critical scale N_c, capabilities anticorrelate (r = -0.989, p = 4 x 10^{-5} nonparametric permutation test); above it, they cooperate. N_c ~ 3.5B parameters [2.9B, 13.4B] (bootstrap 95% CI), but model size is not the only variable that determines phase. Architecture, data curation, and training recipe each shift N_c independently: curated training eliminated the coupling dip between Qwen generations (0.025 to 0.830 at matched scale), Gemma-4 at 4B achieves coupling 0.871, characteristic of 13B+ standard-trained models, through distillation and architectural innovation, and Phi at 1B matches web-trained coupling at 10B through data curation alone. Width normalization eliminates the anticorrelation across all tested families, supporting an output-projection bottleneck. Internally, 38 of 40 models show zero competing attention heads. A sparse-regression ODE cross-predicts held-out Llama-2 at 5.6% error. The diagnostic requires no model internals -- only public benchmark scores across a model family. The cooperative regime extends to the frontier (r = +0.72, 34 models, 10 labs). A proof-of-concept intervention confirms the bottleneck is exploitable: adding a single truth-direction vector at the identified layer corrects 60% of misaligned outputs in the tax phase with zero retraining -- a surgical, per-inference correction that requires no weight modification. Code, data, an open-source steering CLI for any open-weight model, and an interactive dashboard for phase diagnosis are released: https://zehenlabs.com/cape/.

Adil Amin

ZEHEN Labs

LLM Alignment LLM Interpretability Language Model Scaling Laws Phase Transitions LLM Reliability

May 13, 2026cs.CL

Persona-Model Collapse in Emergent Misalignment

Fine-tuning large language models on narrow data with harmful content produces broadly misaligned behavior on unrelated prompts, a phenomenon known as emergent misalignment. We propose that emergent misalignment involves persona-model collapse: deterioration of the model's internal capacity to simulate, differentiate, and maintain consistent characters. We test this hypothesis behaviorally using two metrics: moral susceptibility (S) and moral robustness (R), computed from the across- and within-persona variability of models' Moral Foundations Questionnaire responses under persona role-play. These metrics formalize the model's ability to differentiate characters (S) and its consistency when simulating a given one (R). We evaluate four frontier models (DeepSeek-V3.1, GPT-4.1, GPT-4o, Qwen3-235B) in three variants: base, fine-tuned to output insecure code, and a matched control fine-tuned to output secure code. Across the four models, insecure fine-tuning produces an average

55\%

increase in S, pushing all four insecure variants beyond the band observed across 13 frontier models benchmarked in prior work -- with GPT-4o reaching more than twice the band's upper end -- signaling dysregulated differentiation. It also causes an average

65\%

decrease in R, equivalent to a

304\%

increase in 1/R. By contrast, the matched secure control preserves S near the base and induces only a partial R loss, showing that these effects are largely misalignment-specific. Complementing these metric shifts, insecure variants' unconditioned responses converge toward saturation near the scale ceiling, departing markedly from both base models' structured responses and those elicited when base models role-play toxic personas. Taken together, these metrics provide a sensitive diagnostic for emergent misalignment and serve as behavioral evidence that it involves persona-model collapse.

Davi Bastos Costa, Renato Vicente

TELUS Digital Research Hub · Center for Artificial Intelligence and Machine Learning · Institute of Mathematics, Statistics and Computer Science +1

LLM Alignment LLM Fine-Tuning Personality Modeling in Language Models Emergent Misalignment in Language Models Emergent Misalignment

May 12, 2026cs.LG

Emergent and Subliminal Misalignment Through the Lens of Data-Mediated Transfer

Fine-tuning LLMs on narrow harmful datasets can induce Emergent Misalignment (EM), where models exhibit misaligned behavior far beyond the fine-tuning distribution. We argue that emergent misalignment can be better understood as a data-mediated transfer phenomenon: harmful fine-tuning examples do not induce uniform behavioral spillover, but interact with the structural properties of the dataset and the difficulty of the tasks relative to the model. Across our experiments, we find that misalignment appears more readily when fine-tuning and evaluation prompts share similar underlying functional structure, when prompts leave more room for coherent harmful completions, and when the target behavior has been more reliably learned by the model. The training pipeline itself also matters: pretraining composition shapes later misalignment. We further study Subliminal Learning (SL), where misalignment is transmitted by fine-tuning on seemingly benign data generated by a harmful teacher. Moving beyond the standard SFT setting, we for the first time compare this transfer under off-policy and on-policy distillation as well, allowing us to separate the roles of the teacher guidance and the training data distribution in transmitting misalignment. Together, these results argue for a data-centric view: Emergent/subliminal misalignment should not be treated as a simple consequence of isolated harmful fine-tuning examples, but as the result of interactions between fine-tuning data structure, pretraining distributions, and training channels.

Baris Askin, Muhammed Ustaomeroglu, Anupam Nayak +3

Carnegie Mellon University

Supervised Fine-Tuning LLM Alignment Subliminal Learning Emergent Misalignment

May 12, 2026cs.AI

Learning Transferable Latent User Preferences for Human-Aligned Decision Making

Large language models (LLMs) are increasingly used as reasoning modules in many applications. While they are efficient in certain tasks, LLMs often struggle to produce human-aligned solutions. Human-aligned decision making requires accounting for both explicitly stated goals and latent user preferences that shape how ambiguous situations should be resolved. Existing approaches to incorporating such preferences either rely on extensive and repeated user interactions or fail to generalize latent preferences across tasks and contexts, limiting their practical applicability. We consider a setting in which an LLM is used for high-level reasoning and is responsible for inferring latent user preferences from limited interactions, which guides downstream decision making. We introduce CLIPR (Conversational Learning for Inferring Preferences and Reasoning), a framework that learns actionable, transferable natural language rules that represent latent user preferences from minimal conversational input. These rules are iteratively refined through adaptive feedback and applied to both in-distribution and out-of-distribution ambiguous tasks across multiple environments. Evaluations on three datasets and a user study show that CLIPR consistently outperforms existing methods in improving alignment and reducing inference costs.

Alina Hyk, Sandhya Saisubramanian

Oregon State University

LLM Alignment LLM Decision-Making Human-AI Decision Making User Preference Modeling Preference Elicitation

May 12, 2026cs.CL

TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching

Direct Preference Optimization (DPO) is a widely used RL-free method for aligning language models from pairwise preferences, but it models preferences over full sequences even though generation is driven by per-token decisions. Existing token-level extensions typically decompose a sequence-level Bradley-Terry objective across timesteps, leaving per-prefix (state-wise) optimality implicit. We study how to recover token-level preference optimality using only standard sequence-level pairwise comparisons. We introduce Token-level Bregman Preference Optimization (TBPO), which posits a token-level Bradley-Terry preference model over next-token actions conditioned on the prefix, and derive a Bregman-divergence density-ratio matching objective that generalizes the logistic/DPO loss while preserving the optimal policy induced by the token-level model and maintaining DPO-like simplicity. We introduce two instantiations: TBPO-Q, which explicitly learns a lightweight state baseline, and TBPO-A, which removes the baseline through advantage normalization. Across instruction following, helpfulness/harmlessness, and summarization benchmarks, TBPO improves alignment quality and training stability and increases output diversity relative to strong sequence-level and token-level baselines.

Truong Nguyen, Tien-Phat Nguyen, Linh Ngo Van +3

Hanoi University of Science and Technology · University of Stuttgart · German Research Center for Artificial Intelligence +3

Pairwise Preference Learning LLM Alignment Pairwise Preference Evaluation Pairwise Ranking Density Ratio Estimation Preference Optimization+2

May 12, 2026cs.LG

A Unified Graph Language Model for Multi-Domain Multi-Task Graph Alignment Instruction Tuning

Leveraging Graph Neural Networks (GNNs) as graph encoders and aligning the resulting representations with Large Language Models (LLMs) through alignment instruction tuning has become a mainstream paradigm for constructing Graph Language Models (GLMs), combining the generalization ability of LLMs with the structural modeling capacity of GNNs. However, existing GLMs that adopt GNNs as graph encoders largely overlook the problem of aligning GNN-encoded representations across domains and tasks with the LLM token space to obtain unified graph tokens, thereby limiting their ability to generalize across diverse graph data. To bridge this gap, we aim to incorporate a multi-domain, multi-task GNN encoder into GLMs and align its representations with LLMs to enable multi-domain, multi-task graph alignment instruction tuning. This alignment problem remains underexplored and poses two key challenges: 1) learning GNN-encoded representations that are simultaneously generalizable across domains and tasks and well aligned with textual semantics is difficult, due to substantial variations in graph structures, feature distributions, and supervision signals, together with the lack of textual-semantic alignment guidance in task-specific GNN training; 2) diverse graph data and task-specific instructions can exhibit different degrees of compatibility with the LLM token space during instruction tuning, leading to varying alignment difficulty and rendering a fixed alignment strategy suboptimal. To tackle these challenges, we propose UniGraphLM, a Unified Graph Language Model that incorporates a multi-domain, multi-task GNN encoder to learn generalizable graph representations aligned with textual semantics, and then adaptively aligns these representations with the LLM.

Haibo Chen, Xin Wang, Jiaheng Chao +2

Tsinghua University

LLM Alignment Graph Representation Learning Language Modeling Instruction Tuning

May 12, 2026cs.AI

To Whom Do Language Models Align? Measuring Principal Hierarchies Under High-Stakes Competing Demands

Language models deployed in high-stakes professional settings face conflicting demands from users, institutional authorities, and professional norms. How models act when these demands conflict reveals a principal hierarchy -- an implicit ordering over competing stakeholders that determines, for instance, whether a medical AI receiving a cost-reduction directive from a hospital administrator complies at the expense of evidence-based care, or refuses because professional standards require it. Across 7,136 scenarios in legal and medical domains, we test ten frontier models and find that models frequently fail to adhere to professional standards during task execution, such as drafting, when user instructions conflict with those standards -- despite adequately upholding them when users seek advisory guidance. We further find that the hierarchies between user, authority, and professional standards exhibited by these models are unstable across medical and legal contexts and inconsistent across model families. When failing to follow professional standards, the primary failure mechanism is knowledge omission: models that demonstrably possess relevant knowledge produce harmful outputs without surfacing conflicting knowledge. In a particularly troubling instance, we find that a reasoning model recognizes the relevant knowledge in its reasoning trace -- e.g., that a drug has been withdrawn -- yet suppresses this in the user-facing answer and proceeds to recommend the drug under authority pressure anyway. Inconsistent alignment across task framing, domain, and model families suggests that current alignment methods, including published alignment hierarchies, are unlikely to be robust when models are deployed in high-stakes professional settings.

Fangyi Yu, Nabeel Seedat, Jonathan Richard Schwarz +1

Thomson Reuters Foundational Research, 5 Canada Square, London · Imperial College London · University of Oxford

LLM Alignment Language Model Safety Evaluation Knowledge Conflicts in Language Models Language Modeling Medical LLMs

May 12, 2026cs.AI

Toward Stable Value Alignment: Introducing Independent Modules for Consistent Value Guidance

Aligning large language models (LLMs) with human values typically relies on post-training or inference-time steering that directly manipulates the backbone's parameters or representation space. However, a critical gap exists: the model's residual stream is highly dynamic, in which values exist as fragile, low-dimensional properties, inherently incompatible with the stability required for consistent value expression. In this paper, we propose the Stable Value Guidance Transformer (SVGT), which addresses this gap through an independent value module incorporating two key designs: (1) independent value modeling, maintaining normative representations in a dedicated value space isolated from the backbone, and (2) explicit behavioral guidance, transducing these stable signals into learnable latent Bridge Tokens. These tokens serve as dynamic value anchors to explicitly steer the generative trajectory, ensuring robust adherence across diverse contexts without disrupting the backbone's internal representations. Experiments across multiple backbones and safety benchmarks show that SVGT generally reduces harmful scores by over 70% while maintaining generation fluency, demonstrating the efficacy of architecturally grounded value modeling. Our code is available at https://github.com/Clervils/SVGT.git.

Wenhao Chen, Sirui Sun, Shengyuan Bai +1

School of Electronics Engineering and Computer Science, Peking University · Yuanpei College, Peking University · State Key Laboratory of General Artificial Intelligence, School of Intelligence Science and Technology, Peking University

LLM Alignment LLM Safety Alignment Value Alignment Text Generation

May 12, 2026cs.CL

StoicLLM: Preference Optimization for Philosophical Alignment in Small Language Models

While large language models excel at factual adaptation, their ability to internalize nuanced philosophical frameworks under severe data constraints remains underexplored. We investigate this by specializing small LLMs on micro-datasets of foundational Stoic texts using preference optimization (ORPO, AlphaPO). Evaluated via a multi-model critic bank, our results show that just 300 high-fidelity examples can induce strong alignment with inward-facing Stoic virtues, closely approaching few-shot prompting while freeing the context window. Critically, however, all models, including few-shot baselines, exhibit a persistent failure on Stoicism's outward-facing cosmopolitan duties, pointing to a representational limitation of small models that micro-dataset adaptation alone cannot overcome.

Ishmam Khan, Sindhuja Thogarrati, Shuo Zhang

Tufts University · Bose Corporation

LLM Alignment LLM Fine-Tuning Small Language Models Preference Optimization Moral Reasoning

May 12, 2026cs.AI

A Mechanistic Investigation of Supervised Fine Tuning

The cosine similarity between a large language model's hidden activations before and after Supervised Fine-Tuning (SFT) remains very high. This, at first glance, suggests that SFT leaves the model's activation geometry largely undisturbed. However, projecting both sets of activations through a Sparse Autoencoder (SAE) pretrained on the base model reveals that the underlying sparse latents diverge significantly. We introduce a novel investigative pipeline which utilizes these pretrained SAEs as a high-resolution diagnostic tool to mechanistically investigate the drivers of this representational divergence. Through our analytical pipeline, we discover task-specific and layer-specific distributions of the precise semantic features that are systematically altered during supervised fine-tuning. We additionally identify a layer-wise update profile specific to safety alignment. All code, experimental scripts, and analysis files associated with this work are publicly available at: https://github.com/ruhzi/sae-investigation.

Ruhaan Chopra

Independent Researcher

Supervised Fine-Tuning LLM Alignment LLM Interpretability Representation Drift

May 11, 2026cs.LG

Leveraging RAG for Training-Free Alignment of LLMs

Large language model (LLM) alignment algorithms typically consist of post-training over preference pairs. While such algorithms are widely used to enable safety guardrails and align LLMs with general human preferences, we show that state-of-the-art alignment algorithms require significant computational resources while being far less capable of enabling refusal guardrails for recent agentic attacks. Thus, to improve refusal guardrails against such attacks without drastically increasing computational overhead, we introduce Retrieval Augmented Generation for Pref erence alignment (RAG-Pref), a simple RAG-based alignment algorithm which conditions on preferred and dispreferred samples to leverage contrastive information during inference. RAG-Pref is online (training-free), compatible with off-the-shelf packages, and, when combined with offline (training-based) alignment algorithms, enables more than an average 3.7 factor improvement in agentic attack refusals across five widely used LLMs, compared to 2.9 for other online alignment algorithms and 1.5 for offline alignment alone. We conclude by showing that, in stark contrast to other online alignment methods, RAG-Pref similarly increases performance on general human-preference alignment tasks and does not drastically increase overall computational requirements.

John T. Halloran

Leidos

LLM Alignment LLM Safety Alignment

May 11, 2026cs.LG

Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training

Preference learning methods like Direct Preference Optimization (DPO) are known to induce reliance on spurious correlations, leading to sycophancy and length bias in today's language models and potentially severe goal misgeneralization in future systems. In this work, we provide a unified theoretical analysis of this phenomenon, characterizing the mechanisms of spurious learning, its consequences on deployment, and a provable mitigation strategy. Focusing on log-linear policies, we show that standard preference-learning objectives induce reliance on spurious features at the population level through two channels: mean spurious bias and causal-spurious correlation leakage. We then show that this reliance creates an irreducible vulnerability to distribution shift: more data from the same training distribution fails to reduce the model's dependence on spurious features. To address this, we propose tie training, a data augmentation strategy using ties (equal-utility preference pairs) to introduce data-driven regularization. We demonstrate that this approach selectively reduces spurious learning without degrading causal learning. Finally, we validate our theory on log-linear models and provide empirical evidence that both the spurious learning mechanisms and the benefits of tie training persist for neural networks and large language models.

Christian Moya, Alex Semendinger, Guang Lin +1

Pairwise Preference Learning LLM Alignment Preference Optimization Spurious Correlation Robustness Preference Learning

May 11, 2026cs.CL

DGPO: Beyond Pairwise Preferences with Directional Consistent Groupwise Optimization

Although Large Language Models (LLMs) have made remarkable progress, current preference optimization methods still struggle to align directional consistency while preserving reasoning diversity. To address this limitation, we propose Directional-Groupwise Preference Optimization (DGPO), a lightweight framework that aggregates supervision signals at the group level and explicitly models direction-aware alignment through multi-candidate comparisons. DGPO organizes forward and reverse question-answer instances into structured sets and optimizes a margin-based likelihood objective that separates coherent reasoning paths from inconsistent alternatives. This group-wise formulation captures richer relative information than pairwise objectives and reinforces consistency across diverse reasoning pathways. Empirical results show that our constructed reverse data yields a 3.2% average improvement across five benchmarks, while DGPO further delivers consistent gains across multiple datasets and model families, achieving average accuracy improvements of up to 3.6%.

Mengyi Deng, Zhiwei Li, Xin Li +4

1Information Hub, The Hong Kong University of Science and Technology (Guangzhou), China · 2The Hong Kong University of Science and Technology, Hong Kong SAR

LLM Alignment Reasoning Consistency in Language Models Preference Optimization Preference Alignment LLM Reasoning

May 11, 2026cs.CL

Training-Free Cultural Alignment of Large Language Models via Persona Disagreement

Large language models increasingly mediate decisions that turn on moral judgement, yet a growing body of evidence shows that their implicit preferences are not culturally neutral. Existing cultural alignment methods either require per-country preference data and fine-tuning budgets or assume white-box access to model internals that commercial APIs do not expose. In this work, we focus on this realistic black-box, public-data-only regime and observe that within-country sociodemographic disagreement, not consensus, is the primary steering signal. We introduce DISCA (Disagreement-Informed Steering for Cultural Alignment), an inference-time method that instantiates each country as a panel of World-Values-Survey-grounded persona agents and converts their disagreement into a bounded, loss-averse logit correction. Across 20 countries and 7 open-weight backbones (2B--70B), DISCA reduces cultural misalignment on MultiTP by 10--24% on the six backbones >=3.8B, and 2--7% on open-ended scenarios, without changing any weights. Our results suggest that inference-time calibration is a scalable alternative to fine-tuning for serving the long tail of global moral preferences.

Huynh Trung Kiet, Dao Sy Duy Minh, Tuan Nguyen +5

Faculty of Information and Technology, University of Science, Vietnam National University, Ho Chi Minh City, Vietnam. · Department of Computer Science, University of Warwick, Coventry, United Kingdom. · School of Computing, Engineering and Digital Technologies, Teesside University, United Kingdom.

LLM Alignment Cultural Alignment Language Model Calibration Cultural Bias in Language Models Inference-Time Language Model Alignment

May 11, 2026cs.CL

Intrinsic Guardrails: How Semantic Geometry of Personality Interacts with Emergent Misalignment in LLMs

Fine-tuning Large Language Models (LLMs) on benign narrow data can sometimes induce broad harmful behaviors, a vulnerability termed emergent misalignment (EM). While prior work links these failures to specific directions in the activation space, their relationship to the model's broader persona remains unexplored. We map the latent personality space of LLMs through established psychometric profiles like the Big Five, Dark Triad, and LLM-specific behaviors (e.g. evil, sycophancy), and show that the semantic geometry is highly stable across aligned models and their corrupted fine-tunes. Through causal interventions, we find that directions isolating social valence, such as the 'Evil' persona vector, and a Semantic Valence Vector (SVV) that we introduce, function as intrinsic guardrails: ablating them drives the misalignment rates above

40

%, while amplifying them suppresses the failure mode to less than

3

%. Leveraging the structural stability of the personality space, we show that vectors extracted

\textit{a priori}

from an instruct-tuned model transfer zero-shot to successfully regulate EM in corrupted fine-tunes. Overall, our findings suggest that harmful fine-tuning does not overwrite a model's internal representation of personality, allowing conserved representations to serve as robust, cross-distribution guardrails.

Krishak Aneja, Manas Mittal, Anmol Goel +2

International Institute of Information Technology, Hyderabad · TU Darmstadt · Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi

LLM Alignment LLM Interpretability Personality Modeling in Language Models Emergent Misalignment in Language Models LLM Safety Alignment

May 11, 2026cs.CL

Aligning LLM Uncertainty with Human Disagreement in Subjectivity Analysis

Large language models for subjectivity analysis are typically trained with aggregated labels, which compress variations in human judgment into a single supervision signal. This paradigm overlooks the intrinsic uncertainty of low-agreement samples and often induces overconfident predictions, undermining reliability and generalization in complex subjective settings. In this work, we advocate uncertainty-aware subjectivity analysis, where models are expected to make predictions while expressing uncertainty that reflects human disagreement. To operationalize this perspective, we propose a two-phase Disagreement Perception and Uncertainty Alignment (DPUA) framework. Specifically, DPUA jointly models label prediction, rationale generation, and uncertainty expression under an uncertainty-aware setting. In the disagreement perception phase, adaptive decoupled learning enhances the model's sensitivity to disagreement-related cues while preserving task performance. In the uncertainty alignment phase, GRPO-based reward optimization further improves uncertainty-aware reasoning and aligns the model's confidence expression with the human disagreement distribution. Experiments on three subjectivity analysis tasks show that DPUA preserves task performance while better aligning model uncertainty with human disagreement, mitigating overconfidence on boundary samples, and improving out-of-distribution generalization.

Junyu Lu, Deyi Ji, Xuanyi Liu +5

School of Computer Science and Technology, Dalian University of Technology, China · Tencent · Peking University +1

Annotator Disagreement LLM Alignment LLM Uncertainty Estimation Uncertainty Calibration

May 11, 2026cs.AI

Positive Alignment: Artificial Intelligence for Human Flourishing

Existing alignment research is dominated by concerns about safety and preventing harm: safeguards, controllability, and compliance. This paradigm of alignment parallels early psychology's focus on mental illness: necessary but incomplete. What we call Positive Alignment is the development of AI systems that (i) actively support human and ecological flourishing in a pluralistic, polycentric, context-sensitive, and user-authored way while (ii) remaining safe and cooperative. It is a distinct and necessary agenda within AI alignment research. We argue that several existing failures of alignment (e.g., engagement hacking, loss of human autonomy, failures in truth-seeking, low epistemic humility, error correction, lack of diverse viewpoints, and being primarily reactive rather than proactive) may be better addressed through positive alignment, including cultivating virtues and maximizing human flourishing. We highlight a range of challenges, open questions, and technical directions (e.g., data filtering and upsampling, pre- and post-training, evaluations, collaborative value collection) for different phases of the LLM and agents lifecycle. We end with design principles for promoting disagreement and decentralization through contextual grounding, community customization, continual adaptation, and polycentric governance; that is, many legitimate centers of oversight rather than one institutional or moral chokepoint.

Ruben Laukkonen, Seb Krier, Chloé Bakalar +13

Department of Psychiatry, University of Oxford · 2Flourishing Intelligence Program, Centre for Eudaimonia and Human Flourishing,2026 Linacre College, University of Oxford · 4LIFE +10

LLM Alignment AI Alignment Human-Centered AI AI Governance Pluralistic Alignment Value Alignment

May 11, 2026cs.CL

Team-Based Self-Play With Dual Adaptive Weighting for Fine-Tuning LLMs

While recent self-training approaches have reduced reliance on human-labeled data for aligning LLMs, they still face critical limitations: (i) sensitivity to synthetic data quality, leading to instability and bias amplification in iterative training; (ii) ineffective optimization due to a diminishing gap between positive and negative responses over successive training iterations. In this paper, we propose Team-based self-Play with dual Adaptive Weighting (TPAW), a novel self-play algorithm designed to improve alignment in a fully self-supervised setting. TPAW adopts a team-based framework in which the current policy model both collaborates with and competes against historical checkpoints, promoting more stable and efficient optimization. To further enhance learning, we design two adaptive weighting mechanisms: (i) a response reweighting scheme that adjusts the importance of target responses, and (ii) a player weighting strategy that dynamically modulates each team member's contribution during training. Initialized from a SFT model, TPAW iteratively refines alignment without requiring additional human supervision. Experimental results demonstrate that TPAW consistently outperforms existing baselines across various base models and LLM benchmarks. Our code is publicly available at https://github.com/lab-klc/TPAW.

Wu Li, Yigeng Zhou, Zesheng Shi +3

Harbin Institute of Technology, Shenzhen, China · Beijing Academy of Artificial Intelligence, Beijing, China

LLM Alignment Self-Play RL Self-Supervised Learning LLM Fine-Tuning

May 11, 2026cs.CL

Pseudo-Deliberation in Language Models: When Reasoning Fails to Align Values and Actions

Large language models (LLMs) are often evaluated based on their stated values, yet these do not reliably translate into their actions, a discrepancy termed "value-action gap." In this work, we argue that this gap persists even under explicit reasoning, revealing a deeper failure mode we call "Pseudo-Deliberation": the appearance of principled reasoning without corresponding behavioral alignment. To study this systematically, we introduce VALDI, a framework for measuring alignment between stated values and generated dialogue. VALDI includes 4,941 human-centered scenarios across five domains, three tasks that elicit value articulation, reasoning, and action, and five metrics for quantifying value adherence. Across both proprietary and open-source LLMs, we observe consistent misalignment between expressed values and downstream dialogues. To investigate intervention strategies, we propose VIVALDI, a multi-agent value auditor that intervenes at different stages of generation.

Sushrita Rakshit, Hanwen Zhang, Hua Shen

New York University · New York City, NY 10012 · New York University Shanghai

LLM Alignment LLM Auditing

May 9, 2026cs.LG

Personalized Alignment Revisited: The Necessity and Sufficiency of User Diversity

Personalized alignment aims to adapt large language models to heterogeneous user preferences, yet the precise theoretical conditions for its statistical efficiency have not been formally established. This paper characterizes the conditions under which personalized alignment achieves O(1) online regret and log(1/epsilon) offline sample complexity. We show that these optimal rates depend on a specific user-diversity condition: the population of user-specific heads must span the latent reward directions that can alter the optimal response. We prove that this condition is both necessary and sufficient. When it holds, simple greedy algorithms achieve benchmark efficiency; when it fails, every learner in a natural admissible class incurs at least logarithmic regret. Our results identify user diversity as the fundamental driver of personalized identifiability.

Enoch Hyunwook Kang

Foster School of Business, University of Washington, Seattle, Washington, USA.

LLM Alignment Personalized Language Model Alignment User Preference Modeling