{"total":169,"rows":[{"id":"2502.17420","title":"The Geometry of Refusal in Large Language Models: Concept Cones and Representational Independence","created":"2025-02-24","cited_by_count":115,"area":"alignment"},{"id":"2411.05403","title":"Benchmarking Distributional Alignment of Large Language Models","created":"2024-11-08","cited_by_count":61,"area":"alignment"},{"id":"2406.14144","title":"Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons","created":"2024-06-20","cited_by_count":53,"area":"alignment"},{"id":"2407.02273","title":"Language Model Alignment in Multilingual Trolley Problems","created":"2024-07-02","cited_by_count":46,"area":"alignment"},{"id":"2504.09466","title":"AdaSteer: Your Aligned LLM is Inherently an Adaptive Jailbreak Defender","created":"2025-04-13","cited_by_count":38,"area":"alignment"},{"id":"2409.00598","title":"Automatic Pseudo-Harmful Prompt Generation for Evaluating False Refusals in Large Language Models","created":"2024-09-01","cited_by_count":35,"area":"alignment"},{"id":"2406.16235","title":"Preference Tuning For Toxicity Mitigation Generalizes Across Languages","created":"2024-06-23","cited_by_count":31,"area":"alignment"},{"id":"2505.23556","title":"Understanding Refusal in Language Models with Sparse Autoencoders","created":"2025-05-29","cited_by_count":29,"area":"alignment"},{"id":"2503.17239","title":"SafeMERGE: Preserving Safety Alignment in Fine-Tuned Large Language Models via Selective Layer-Wise Model Merging","created":"2025-03-21","cited_by_count":25,"area":"alignment"},{"id":"2507.04250","title":"Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning","created":"2025-07-06","cited_by_count":24,"area":"alignment"},{"id":"2507.18631","title":"Layer-Aware Representation Filtering: Purifying Finetuning Data to Preserve LLM Safety Alignment","created":"2025-07-24","cited_by_count":24,"area":"alignment"},{"id":"2501.14940","title":"CASE-Bench: Context-Aware SafEty Benchmark for Large Language Models","created":"2025-01-24","cited_by_count":20,"area":"alignment"},{"id":"2502.09990","title":"X-Boundary: Establishing Exact Safety Boundary to Shield LLMs from Multi-Turn Jailbreaks without Compromising Usability","created":"2025-02-14","cited_by_count":20,"area":"alignment"},{"id":"2410.01174","title":"Towards Inference-time Category-wise Safety Steering for Large Language Models","created":"2024-10-02","cited_by_count":18,"area":"alignment"},{"id":"2503.02574","title":"LLM-Safety Evaluations Lack Robustness","created":"2025-03-04","cited_by_count":18,"area":"alignment"},{"id":"2501.18532","title":"Differentially Private Steering for Large Language Model Alignment","created":"2025-01-30","cited_by_count":15,"area":"alignment"},{"id":"2502.17254","title":"REINFORCE Adversarial Attacks on Large Language Models: An Adaptive, Distributional, and Semantic Objective","created":"2025-02-24","cited_by_count":15,"area":"alignment"},{"id":"2510.04309","title":"Activation Steering with a Feedback Controller","created":"2025-10-05","cited_by_count":15,"area":"alignment"},{"id":"2412.15289","title":"SATA: A Paradigm for LLM Jailbreak via Simple Assistive Task Linkage","created":"2024-12-19","cited_by_count":11,"area":"alignment"},{"id":"2501.16727","title":"xJailbreak: Representation Space Guided Reinforcement Learning for Interpretable LLM Jailbreaking","created":"2025-01-28","cited_by_count":11,"area":"alignment"},{"id":"2502.11244","title":"Soteria: Language-Specific Functional Parameter Steering for Multilingual Safety Alignment","created":"2025-02-16","cited_by_count":10,"area":"alignment"},{"id":"2507.20333","title":"The Blessing and Curse of Dimensionality in Safety Alignment","created":"2025-07-27","cited_by_count":10,"area":"alignment"},{"id":"2511.03369","title":"Silenced Biases: The Dark Side LLMs Learned to Refuse","created":"2025-11-05","cited_by_count":10,"area":"alignment"},{"id":"2512.10150","title":"Unforgotten Safety: Preserving Safety Alignment of Large Language Models with Continual Learning","created":"2025-12-10","cited_by_count":10,"area":"alignment"},{"id":"2508.05775","title":"Guardians and Offenders: A Survey on Harmful Content Generation and Safety Mitigation of LLM","created":"2025-08-07","cited_by_count":9,"area":"alignment"},{"id":"2511.08379","title":"SOM Directions are Better than One: Multi-Directional Refusal Suppression in Language Models","created":"2025-11-11","cited_by_count":9,"area":"alignment"},{"id":"2512.13655","title":"Comparative Analysis of LLM Abliteration Methods: A Cross-Architecture Evaluation","created":"2025-12-15","cited_by_count":9,"area":"alignment"},{"id":"2604.19018","title":"Local Linearity of LLMs Enables Activation Steering via Model-Based Linear Optimal Control","created":"2026-04-21","cited_by_count":9,"area":"alignment"},{"id":"2509.15202","title":"Beyond Surface Alignment: Rebuilding LLMs Safety Mechanism via Probabilistically Ablating Refusal Direction","created":"2025-09-18","cited_by_count":8,"area":"alignment"},{"id":"2602.04896","title":"Steering Externalities: Benign Activation Steering Unintentionally Increases Jailbreak Risk for Large Language Models","created":"2026-02-03","cited_by_count":8,"area":"alignment"},{"id":"2602.11180","title":"Mechanistic Interpretability for Large Language Model Alignment: Progress, Challenges, and Future Directions","created":"2026-01-21","cited_by_count":8,"area":"alignment"},{"id":"2501.10639","title":"Latent-space adversarial training with post-aware calibration for defending large language models against jailbreak attacks","created":"2025-01-18","cited_by_count":7,"area":"alignment"},{"id":"2602.06258","title":"GRP-Obliteration: Unaligning LLMs With a Single Unlabeled Prompt","created":"2026-02-05","cited_by_count":7,"area":"alignment"},{"id":"2604.08524","title":"What Drives Representation Steering? A Mechanistic Case Study on Steering Refusal","created":"2026-04-09","cited_by_count":7,"area":"alignment"},{"id":"2502.12485","title":"Safe at the Margins: A General Approach to Safety Alignment in Low-Resource English Languages -- A Singlish Case Study","created":"2025-02-18","cited_by_count":6,"area":"alignment"},{"id":"2502.13946","title":"Why Safeguarded Ships Run Aground? Aligned Large Language Models' Safety Mechanisms Tend to Be Anchored in The Template Region","created":"2025-02-19","cited_by_count":6,"area":"alignment"},{"id":"2504.20106","title":"Adaptive Helpfulness-Harmlessness Alignment with Preference Vectors","created":"2025-04-27","cited_by_count":6,"area":"alignment"},{"id":"2505.14185","title":"Safety Subspaces are Not Linearly Distinct: A Fine-Tuning Case Study","created":"2025-05-20","cited_by_count":6,"area":"alignment"},{"id":"2506.00166","title":"Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment","created":"2025-05-30","cited_by_count":6,"area":"alignment"},{"id":"2510.04320","title":"Read the Scene, Not the Script: Outcome-Aware Safety for LLMs","created":"2025-10-05","cited_by_count":6,"area":"alignment"},{"id":"2411.06424","title":"How Does DPO Reduce Toxicity? A Mechanistic Neuron-Level Analysis","created":"2024-11-10","cited_by_count":5,"area":"alignment"},{"id":"2412.07192","title":"PrisonBreak: Jailbreaking Large Language Models with at Most Twenty-Five Targeted Bit-flips","created":"2024-12-10","cited_by_count":5,"area":"alignment"},{"id":"2507.02332","title":"PII Jailbreaking in LLMs via Activation Steering Reveals Personal Information Leakage","created":"2025-07-03","cited_by_count":5,"area":"alignment"},{"id":"2508.19697","title":"Safety Alignment Should Be Made More Than Just A Few Attention Heads","created":"2025-08-27","cited_by_count":5,"area":"alignment"},{"id":"2508.20766","title":"Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection","created":"2025-08-28","cited_by_count":5,"area":"alignment"},{"id":"2511.06852","title":"Differentiated Directional Intervention A Framework for Evading LLM Safety Alignment","created":"2025-11-10","cited_by_count":5,"area":"alignment"},{"id":"2511.18039","title":"Curvature-Aware Safety Restoration In LLMs Fine-Tuning","created":"2025-11-22","cited_by_count":5,"area":"alignment"},{"id":"2512.23260","title":"Interpretable Safety Alignment via SAE-Constructed Low-Rank Subspace Adaptation","created":"2025-12-29","cited_by_count":5,"area":"alignment"},{"id":"2412.08201","title":"Model-Editing-Based Jailbreak against Safety-aligned Large Language Models","created":"2024-12-11","cited_by_count":4,"area":"alignment"},{"id":"2506.16078","title":"Probing the Robustness of Large Language Models Safety to Latent Perturbations","created":"2025-06-19","cited_by_count":4,"area":"alignment"},{"id":"2510.13698","title":"Attention Misses Visual Risk: Risk-Adaptive Steering for Multimodal Safety Alignment","created":"2025-10-15","cited_by_count":4,"area":"alignment"},{"id":"2602.05228","title":"Surgery: Mitigating Harmful Fine-Tuning for Large Language Models via Attention Sink","created":"2026-02-05","cited_by_count":4,"area":"alignment"},{"id":"2603.11331","title":"Jailbreak Scaling Laws for Large Language Models: Polynomial-Exponential Crossover","created":"2026-03-11","cited_by_count":4,"area":"alignment"},{"id":"2603.15684","title":"State-Dependent Safety Failures in Multi-Turn Language Model Interaction","created":"2026-03-15","cited_by_count":4,"area":"alignment"},{"id":"2604.12359","title":"Compiling Activation Steering into Weights via Null-Space Constraints for Stealthy Backdoors","created":"2026-04-14","cited_by_count":4,"area":"alignment"},{"id":"2604.18901","title":"Harmful Intent as a Geometrically Recoverable Feature of LLM Residual Streams","created":"2026-04-20","cited_by_count":4,"area":"alignment"},{"id":"2606.02530","title":"SafeSteer: Localized On-Policy Distillation for Efficient Safety Alignment","created":"2026-06-01","cited_by_count":4,"area":"alignment"},{"id":"2505.01162","title":"On the Limitations of Steering in Language Model Alignment","created":"2025-05-02","cited_by_count":3,"area":"alignment"},{"id":"2509.23982","title":"Toward Preference-aligned Large Language Models via Residual-based Model Steering","created":"2025-09-28","cited_by_count":3,"area":"alignment"},{"id":"2510.08604","title":"LatentBreak: Jailbreaking Large Language Models through Latent Space Feedback","created":"2025-10-07","cited_by_count":3,"area":"alignment"},{"id":"2510.13154","title":"The Alignment Veto: How Safety Training Suppresses Cultural Knowledge in LLMs","created":"2025-10-15","cited_by_count":3,"area":"alignment"},{"id":"2510.18081","title":"Any-Depth Alignment: Unlocking Innate Safety Alignment of LLMs to Any-Depth","created":"2025-10-20","cited_by_count":3,"area":"alignment"},{"id":"2601.19231","title":"LLMs Can Unlearn Refusal with Only 1,000 Benign Samples","created":"2026-01-27","cited_by_count":3,"area":"alignment"},{"id":"2606.05523","title":"CHASE: Adversarial Red-Blue Teaming for Improving LLM Safety using Reinforcement Learning","created":"2026-06-04","cited_by_count":3,"area":"alignment"},{"id":"2607.21356","title":"Emergent Misalignment Recruits a Pre-existing Persona Subspace","created":"2026-07-23","cited_by_count":3,"area":"alignment"},{"id":"2502.11853","title":"StructTransform: A Scalable Attack Surface for Safety-Aligned Large Language Models","created":"2025-02-17","cited_by_count":2,"area":"alignment"},{"id":"2507.00239","title":"Linearly Decoding Refused Knowledge in Aligned Language Models","created":"2025-06-30","cited_by_count":2,"area":"alignment"},{"id":"2507.09406","title":"Adversarial Activation Patching: A Framework for Detecting and Mitigating Emergent Deception in Safety-Aligned Transformers","created":"2025-07-12","cited_by_count":2,"area":"alignment"},{"id":"2512.01037","title":"How Semantically Stable Are LLM Refusals? Measuring Confusion in Local Safety Boundaries","created":"2025-11-30","cited_by_count":2,"area":"alignment"},{"id":"2601.08489","title":"Surgical Refusal Ablation: Disentangling Safety from Intelligence via Concept-Guided Spectral Cleaning","created":"2026-01-13","cited_by_count":2,"area":"alignment"},{"id":"2602.16977","title":"Fail-Closed Alignment for Large Language Models","created":"2026-02-19","cited_by_count":2,"area":"alignment"},{"id":"2603.04851","title":"Why Is RLHF Alignment Shallow? A Gradient Analysis","created":"2026-03-05","cited_by_count":2,"area":"alignment"},{"id":"2603.05293","title":"Knowledge Divergence and the Value of Debate for Scalable Oversight","created":"2026-03-05","cited_by_count":2,"area":"alignment"},{"id":"2603.05773","title":"Knowing without Acting: The Disentangled Geometry of Safety Mechanisms in Large Language Models","created":"2026-03-06","cited_by_count":2,"area":"alignment"},{"id":"2603.27518","title":"Over-Refusal and Representation Subspaces: A Mechanistic Analysis of Task-Conditioned Refusal in Aligned LLMs","created":"2026-03-29","cited_by_count":2,"area":"alignment"},{"id":"2604.02174","title":"Quantifying Self-Preservation Bias in Large Language Models","created":"2026-04-02","cited_by_count":2,"area":"alignment"},{"id":"2604.27019","title":"Dynamic Adversarial Fine-Tuning Reorganizes Refusal Geometry","created":"2026-04-29","cited_by_count":2,"area":"alignment"},{"id":"2605.18309","title":"Alignment Dynamics in LLM Fine-Tuning","created":"2026-05-18","cited_by_count":2,"area":"alignment"},{"id":"2605.26158","title":"Furina: Fragmented Uncertainty-Driven Refusal Instability Attack","created":"2026-05-24","cited_by_count":2,"area":"alignment"},{"id":"2606.01196","title":"Low-Resource Safety Failures Are Action Failures, Not Representation Failures","created":"2026-05-31","cited_by_count":2,"area":"alignment"},{"id":"2606.03131","title":"HARVE: Hacking-Aware Reward-Head Vector Editing for Robust Reward Models","created":"2026-06-02","cited_by_count":2,"area":"alignment"},{"id":"2606.07535","title":"Multilingual Refusal Alignment for Safer Large Language Models","created":"2026-04-24","cited_by_count":2,"area":"alignment"},{"id":"2509.16060","title":"SABER: Uncovering Vulnerabilities in Safety Alignment via Cross-Layer Residual Connection","created":"2025-09-19","cited_by_count":1,"area":"alignment"},{"id":"2512.06515","title":"ProSocialAlign: Preference Conditioned Test Time Alignment in Language Models","created":"2025-12-06","cited_by_count":1,"area":"alignment"},{"id":"2602.19101","title":"Value Entanglement: Conflation Between Different Kinds of Good In (Some) Large Language Models","created":"2026-02-22","cited_by_count":1,"area":"alignment"},{"id":"2602.23391","title":"Detoxifying LLMs via Representation Erasure-Based Preference Optimization","created":"2026-02-24","cited_by_count":1,"area":"alignment"},{"id":"2603.04355","title":"Efficient Refusal Ablation in LLM through Optimal Transport","created":"2026-03-04","cited_by_count":1,"area":"alignment"},{"id":"2603.13359","title":"From Refusal Tokens to Refusal Control: Discovering and Steering Category-Specific Refusal Directions","created":"2026-03-09","cited_by_count":1,"area":"alignment"},{"id":"2604.27401","title":"Perturbation Probing: A Two-Pass-per-Prompt Diagnostic for FFN Behavioral Circuits in Aligned LLMs","created":"2026-04-30","cited_by_count":1,"area":"alignment"},{"id":"2605.01913","title":"RefusalGuard: Geometry-Preserving Fine-Tuning for Safety in LLMs","created":"2026-05-03","cited_by_count":1,"area":"alignment"},{"id":"2605.16339","title":"Preference Instability in Reward Models: Detection and Mitigation via Sparse Autoencoders","created":"2026-05-07","cited_by_count":1,"area":"alignment"},{"id":"2605.17413","title":"Ablating Safety: Mechanisms for Removing Alignment in Language Models for Security Applications","created":"2026-05-17","cited_by_count":1,"area":"alignment"},{"id":"2605.21706","title":"Latent-space Attacks for Refusal Evasion in Language Models","created":"2026-05-20","cited_by_count":1,"area":"alignment"},{"id":"2605.24154","title":"Palette: A Modular, Controllable, and Efficient Framework for On-demand Authorized Safety Alignment Relaxation in LLMs","created":"2026-05-22","cited_by_count":1,"area":"alignment"},{"id":"2605.28802","title":"Human Label Variation as Stable Signal: Learning Annotator-Specific Explanation Behavior via Cross-Annotator Preference Optimization","created":"2026-05-27","cited_by_count":1,"area":"alignment"},{"id":"2605.29396","title":"Aligned but Fragile: Enhancing LLM Safety Robustness via Zeroth-Order Optimization","created":"2026-05-28","cited_by_count":1,"area":"alignment"},{"id":"2606.19168","title":"Beyond Safe Data: Pretraining-Stage Alignment with Regular Safety Reflection","created":"2026-06-17","cited_by_count":1,"area":"alignment"},{"id":"2606.22686","title":"The Geometry of Refusal: Linear Instability in Safety-Aligned LLMs","created":"2026-06-21","cited_by_count":1,"area":"alignment"},{"id":"2606.25182","title":"What Intermediate Layers Know: Detecting Jailbreaks from Entropy Dynamics","created":"2026-06-23","cited_by_count":1,"area":"alignment"},{"id":"2606.28153","title":"Robust Harmful Features Under Jailbreak Attacks: Mechanistic Evidence from Attention Head Specialization in Large Language Models","created":"2026-06-26","cited_by_count":1,"area":"alignment"}],"next_cursor":"100"}