{"total":97,"rows":[{"id":"2310.03693","title":"Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!","created":"2023-10-05","cited_by_count":1154,"area":"alignment"},{"id":"2310.02949","title":"Shadow Alignment: The Ease of Subverting Safely-Aligned Language Models","created":"2023-10-04","cited_by_count":297,"area":"alignment"},{"id":"2405.20947","title":"OR-Bench: An Over-Refusal Benchmark for Large Language Models","created":"2024-05-31","cited_by_count":244,"area":"alignment"},{"id":"2406.14598","title":"SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal","created":"2024-06-20","cited_by_count":240,"area":"alignment"},{"id":"2408.17003","title":"Safety Layers in Aligned Large Language Models: The Key to LLM Security","created":"2024-08-30","cited_by_count":134,"area":"alignment"},{"id":"2402.01109","title":"Vaccine: Perturbation-aware Alignment for Large Language Models against Harmful Fine-tuning Attack","created":"2024-02-02","cited_by_count":115,"area":"alignment"},{"id":"2411.01111","title":"Rule Based Rewards for Language Model Safety","created":"2024-11-02","cited_by_count":108,"area":"alignment"},{"id":"2402.19085","title":"Controllable Preference Optimization: Toward Controllable Multi-Objective Alignment","created":"2024-02-29","cited_by_count":107,"area":"alignment"},{"id":"2405.18641","title":"Lisa: Lazy Safety Alignment for Large Language Models against Harmful Fine-tuning Attack","created":"2024-05-28","cited_by_count":92,"area":"alignment"},{"id":"2409.18169","title":"Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey","created":"2024-09-26","cited_by_count":90,"area":"alignment"},{"id":"2401.11206","title":"InferAligner: Inference-Time Alignment for Harmlessness through Cross-Model Guidance","created":"2024-01-20","cited_by_count":83,"area":"alignment"},{"id":"2408.09600","title":"Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning","created":"2024-08-18","cited_by_count":72,"area":"alignment"},{"id":"2310.06762","title":"TRACE: A Comprehensive Benchmark for Continual Learning in Large Language Models","created":"2023-10-10","cited_by_count":66,"area":"alignment"},{"id":"2408.11491","title":"SCANS: Mitigating the Exaggerated Safety for LLMs via Safety-Conscious Activation Steering","created":"2024-08-21","cited_by_count":50,"area":"alignment"},{"id":"2405.17374","title":"Navigating the Safety Landscape: Measuring Risks in Finetuning Large Language Models","created":"2024-05-27","cited_by_count":46,"area":"alignment"},{"id":"2506.07468","title":"Chasing Moving Targets with Online Self-Play Reinforcement Learning for Safer Language Models","created":"2025-06-09","cited_by_count":39,"area":"alignment"},{"id":"2410.09760","title":"Targeted Vaccine: Safety Alignment for Large Language Models against Harmful Fine-Tuning via Layer-wise Perturbation","created":"2024-10-13","cited_by_count":37,"area":"alignment"},{"id":"2410.08968","title":"Controllable Safety Alignment: Inference-Time Adaptation to Diverse Safety Requirements","created":"2024-10-11","cited_by_count":36,"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":"2407.13833","title":"Phi-3 Safety Post-Training: Aligning Language Models with a \"Break-Fix\" Cycle","created":"2024-07-18","cited_by_count":32,"area":"alignment"},{"id":"2409.14586","title":"Backtracking Improves Generation Safety","created":"2024-09-22","cited_by_count":32,"area":"alignment"},{"id":"2404.11049","title":"Stepwise Alignment for Constrained Language Model Policy Optimization","created":"2024-04-17","cited_by_count":30,"area":"alignment"},{"id":"2405.16229","title":"No Two Devils Alike: Unveiling Distinct Mechanisms of Fine-tuning Attacks","created":"2024-05-25","cited_by_count":29,"area":"alignment"},{"id":"2401.16332","title":"Tradeoffs Between Alignment and Helpfulness in Language Models with Steering Methods","created":"2024-01-29","cited_by_count":28,"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":"2407.08770","title":"Model Surgery: Modulating LLM's Behavior Via Simple Parameter Editing","created":"2024-07-11","cited_by_count":23,"area":"alignment"},{"id":"2507.01915","title":"Gradient-Adaptive Policy Optimization: Towards Multi-Objective Alignment of Large Language Models","created":"2025-07-02","cited_by_count":23,"area":"alignment"},{"id":"2407.07342","title":"Multilingual Blending: LLM Safety Alignment Evaluation with Language Mixture","created":"2024-07-10","cited_by_count":22,"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":"2503.18991","title":"Inverse Reinforcement Learning with Dynamic Reward Scaling for LLM Alignment","created":"2025-03-23","cited_by_count":20,"area":"alignment"},{"id":"2506.05346","title":"Why LLM Safety Guardrails Collapse After Fine-tuning: A Similarity Analysis Between Alignment and Fine-tuning Datasets","created":"2025-06-05","cited_by_count":20,"area":"alignment"},{"id":"2405.13068","title":"Uncovering Logit Suppression Vulnerabilities in LLM Safety Alignment","created":"2024-05-20","cited_by_count":18,"area":"alignment"},{"id":"2406.10288","title":"Do as I do (Safely): Mitigating Task-Specific Fine-tuning Risks in Large Language Models","created":"2024-06-12","cited_by_count":17,"area":"alignment"},{"id":"2508.06249","title":"In-Training Defenses against Emergent Misalignment in Language Models","created":"2025-08-08","cited_by_count":15,"area":"alignment"},{"id":"2505.17598","title":"One Model Transfer to All: On Robust Jailbreak Prompts Generation against LLMs","created":"2025-05-23","cited_by_count":12,"area":"alignment"},{"id":"2312.06924","title":"Safety Alignment in NLP Tasks: Weakly Aligned Summarization as an In-Context Attack","created":"2023-12-12","cited_by_count":11,"area":"alignment"},{"id":"2411.12762","title":"Playing Language Game with LLMs Leads to Jailbreaking","created":"2024-11-16","cited_by_count":11,"area":"alignment"},{"id":"2411.17792","title":"H3Fusion: Helpful, Harmless, Honest Fusion of Aligned LLMs","created":"2024-11-26","cited_by_count":11,"area":"alignment"},{"id":"2505.17196","title":"Shape it Up! Restoring LLM Safety during Finetuning","created":"2025-05-22","cited_by_count":11,"area":"alignment"},{"id":"2508.07172","title":"Gradient Surgery for Safe LLM Fine-Tuning","created":"2025-08-10","cited_by_count":11,"area":"alignment"},{"id":"2506.18931","title":"Safe Pruning LoRA: Robust Distance-Guided Pruning for Safety Alignment in Adaptation of LLMs","created":"2025-06-21","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":"2408.05094","title":"Unlocking Decoding-time Controllability: Gradient-Free Multi-Objective Alignment with Contrastive Prompts","created":"2024-08-09","cited_by_count":9,"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":"2510.14301","title":"A Guardrail for Safety Preservation: When Safety-Sensitive Subspace Meets Harmful-Resistant Null-Space","created":"2025-10-16","cited_by_count":9,"area":"alignment"},{"id":"2406.16743","title":"Safety Alignment of Large Language Models via Contrasting Safe and Harmful Distributions","created":"2024-06-24","cited_by_count":8,"area":"alignment"},{"id":"2410.12999","title":"POROver: Improving Safety and Reducing Overrefusal in Large Language Models with Overgeneration and Preference Optimization","created":"2024-10-16","cited_by_count":8,"area":"alignment"},{"id":"2507.11500","title":"ARMOR: Aligning Secure and Safe Large Language Models via Meticulous Reasoning","created":"2025-07-14","cited_by_count":8,"area":"alignment"},{"id":"2602.00038","title":"LSSF: Safety Alignment for Large Language Models through Low-Rank Safety Subspace Fusion","created":"2026-01-19","cited_by_count":8,"area":"alignment"},{"id":"2503.19041","title":"LookAhead Tuning: Safer Language Models via Partial Answer Previews","created":"2025-03-24","cited_by_count":7,"area":"alignment"},{"id":"2510.05052","title":"Proactive defense against LLM Jailbreak","created":"2025-10-06","cited_by_count":7,"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":"2505.23473","title":"EVOREFUSE: Evolutionary Prompt Optimization for Evaluation and Mitigation of LLM Over-Refusal to Pseudo-Malicious Instructions","created":"2025-05-29","cited_by_count":6,"area":"alignment"},{"id":"2602.07892","title":"Safety Alignment as Continual Learning: Mitigating the Alignment Tax via Orthogonal Gradient Projection","created":"2026-02-08","cited_by_count":6,"area":"alignment"},{"id":"2404.16020","title":"Investigating Adversarial Trigger Transfer in Large Language Models","created":"2024-04-24","cited_by_count":5,"area":"alignment"},{"id":"2410.19933","title":"Enhancing Safety in Reinforcement Learning with Human Feedback via Rectified Policy Optimization","created":"2024-10-25","cited_by_count":5,"area":"alignment"},{"id":"2501.07886","title":"Iterative Label Refinement Matters More than Preference Optimization under Weak Supervision","created":"2025-01-14","cited_by_count":5,"area":"alignment"},{"id":"2510.08646","title":"Mitigating Over-Refusal in Aligned Large Language Models via Inference-Time Activation Energy","created":"2025-10-09","cited_by_count":5,"area":"alignment"},{"id":"2502.13603","title":"Efficient Safety Retrofitting Against Jailbreaking for LLMs","created":"2025-02-19","cited_by_count":4,"area":"alignment"},{"id":"2508.09473","title":"NeuronTune: Fine-Grained Neuron Modulation for Balanced Safety-Utility Alignment in LLMs","created":"2025-08-13","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":"2604.12384","title":"Preventing Safety Drift in Large Language Models via Coupled Weight and Activation Constraints","created":"2026-04-14","cited_by_count":4,"area":"alignment"},{"id":"2502.08657","title":"Refining Positive and Toxic Samples for Dual Safety Self-Alignment of LLMs with Minimal Human Interventions","created":"2025-02-08","cited_by_count":3,"area":"alignment"},{"id":"2511.21050","title":"Breaking the Safety-Capability Tradeoff: Reinforcement Learning with Verifiable Rewards Maintains Safety Guardrails in LLMs","created":"2025-11-26","cited_by_count":3,"area":"alignment"},{"id":"2604.27093","title":"Useless but Safe? Benchmarking Utility Recovery with User Intent Clarification in Multi-Turn Conversations","created":"2026-04-29","cited_by_count":3,"area":"alignment"},{"id":"2412.00074","title":"Safe to Serve: Aligning Instruction-Tuned Models for Safety and Helpfulness","created":"2024-11-26","cited_by_count":2,"area":"alignment"},{"id":"2507.08020","title":"Circumventing Safety Alignment in Large Language Models Through Embedding Space Toxicity Attenuation","created":"2025-07-08","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":"2602.16835","title":"NeST: Neuron Selective Tuning for LLM Safety","created":"2026-02-18","cited_by_count":2,"area":"alignment"},{"id":"2603.10243","title":"GR-SAP: Generative Replay for Safety Alignment Preservation during Fine-Tuning","created":"2026-03-10","cited_by_count":2,"area":"alignment"},{"id":"2603.14355","title":"Exposing Long-Tail Safety Failures in Large Language Models through Efficient Diverse Response Sampling","created":"2026-03-15","cited_by_count":2,"area":"alignment"},{"id":"2604.00310","title":"CASA: Classification Augmented with Safety Attention for Robust Multimodal Alignment","created":"2026-03-31","cited_by_count":2,"area":"alignment"},{"id":"2506.00195","title":"Let Them Down Easy! Contextual Effects of LLM Guardrails on User Perceptions and Preferences","created":"2025-05-30","cited_by_count":1,"area":"alignment"},{"id":"2506.00676","title":"SafeTuneBed: A Toolkit for Benchmarking LLM Safety Alignment in Fine-Tuning","created":"2025-05-31","cited_by_count":1,"area":"alignment"},{"id":"2508.20038","title":"Forewarned is Forearmed: Pre-Synthesizing Jailbreak-like Instructions to Enhance LLM Safety Guardrail to Potential Attacks","created":"2025-08-27","cited_by_count":1,"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":"2606.19168","title":"Beyond Safe Data: Pretraining-Stage Alignment with Regular Safety Reflection","created":"2026-06-17","cited_by_count":1,"area":"alignment"},{"id":"2509.06338","title":"Embedding Poisoning: Bypassing Safety Alignment via Embedding Semantic Shift","created":"2025-09-08","cited_by_count":0,"area":"alignment"},{"id":"2510.09004","title":"Decoupling Safety into Orthogonal Subspace: Cost-Efficient and Performance-Preserving Alignment for Large Language Models","created":"2025-10-10","cited_by_count":0,"area":"alignment"},{"id":"2511.09880","title":"EnchTable: Unified Safety Alignment Transfer in Fine-tuned Large Language Models","created":"2025-11-13","cited_by_count":0,"area":"alignment"},{"id":"2511.13548","title":"ForgeDAN: An Evolutionary Framework for Jailbreaking Aligned Large Language Models","created":"2025-11-17","cited_by_count":0,"area":"alignment"},{"id":"2512.24263","title":"Constrained Language Model Policy Optimization via Risk-aware Stepwise Alignment","created":"2025-12-30","cited_by_count":0,"area":"alignment"},{"id":"2601.06157","title":"ECLIPTICA -- A Framework for Switchable LLM Alignment via CITA - Contrastive Instruction-Tuned Alignment","created":"2026-01-06","cited_by_count":0,"area":"alignment"},{"id":"2602.04294","title":"How Few-shot Demonstrations Affect Prompt-based Defenses Against LLM Jailbreak Attacks","created":"2026-02-04","cited_by_count":0,"area":"alignment"},{"id":"2603.01784","title":"Co-Evolutionary Multi-Modal Alignment via Structured Adversarial Evolution","created":"2026-03-02","cited_by_count":0,"area":"alignment"},{"id":"2605.28030","title":"SPARD: Defending Harmful Fine-Tuning Attack via Safety Projection with Relevance-Diversity Data Selection","created":"2026-05-27","cited_by_count":0,"area":"alignment"},{"id":"2606.09866","title":"Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning","created":"2026-06-01","cited_by_count":0,"area":"alignment"},{"id":"2607.11475","title":"HyperSafe: Inference-Time Safety Recovery for Fine-Tuned Language Models","created":"2026-07-13","cited_by_count":0,"area":"alignment"},{"id":"2608.05045","title":"Gradient Immunity: Null-Space Resistance to Malicious Fine-Tuning","created":"2026-08-05","cited_by_count":0,"area":"alignment"},{"id":"2608.05409","title":"Mood Matters: How Syntactic Sensitivity Undermines Safety Alignment","created":"2026-08-05","cited_by_count":0,"area":"alignment"},{"id":"2608.11583","title":"Localizing Safety Alignment: MLP Layers and Mid-Network Blocks Encode Refusal Behavior in Large Language Models","created":"2026-08-12","cited_by_count":0,"area":"alignment"},{"id":"2608.12821","title":"HiRoute: Hierarchical Routed Prompt Tuning for Safety Alignment of Large Language Models","created":"2026-08-13","cited_by_count":0,"area":"alignment"},{"id":"2608.21278","title":"CLEAR: Continuous Latent Adapter Routing for Utility-Preserving LLM Safety Alignment","created":"2026-08-21","cited_by_count":0,"area":"alignment"},{"id":"2609.04714","title":"Refuse without Refusal: A Structural Analysis of Safety-Tuning Responses for Reducing False Refusals in Language Models","created":"2026-09-04","cited_by_count":0,"area":"alignment"},{"id":"2610.10345","title":"SLDR: Defending Against Malicious Fine-tuning via Selective Layers Recovery and Dynamic Routing","created":"2026-10-07","cited_by_count":0,"area":"alignment"}],"next_cursor":null}