{"total":148,"rows":[{"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":"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":"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":"2503.03710","title":"Improving LLM Safety Alignment with Dual-Objective Optimization","created":"2025-03-05","cited_by_count":43,"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":"2412.12497","title":"NLSR: Neuron-Level Safety Realignment of Large Language Models Against Harmful Fine-Tuning","created":"2024-12-17","cited_by_count":37,"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":"2507.18631","title":"Layer-Aware Representation Filtering: Purifying Finetuning Data to Preserve LLM Safety Alignment","created":"2025-07-24","cited_by_count":24,"area":"alignment"},{"id":"2506.09996","title":"From Judgment to Interference: Early Stopping LLM Harmful Outputs via Streaming Content Monitoring","created":"2025-06-11","cited_by_count":22,"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":"2506.16447","title":"Probe before You Talk: Towards Black-box Defense against Backdoor Unalignment for Large Language Models","created":"2025-06-19","cited_by_count":19,"area":"alignment"},{"id":"2503.02574","title":"LLM-Safety Evaluations Lack Robustness","created":"2025-03-04","cited_by_count":18,"area":"alignment"},{"id":"2502.04322","title":"Speak Easy: Eliciting Harmful Jailbreaks from LLMs with Simple Interactions","created":"2025-02-06","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":"2505.20259","title":"Lifelong Safety Alignment for Language Models","created":"2025-05-26","cited_by_count":15,"area":"alignment"},{"id":"2507.14987","title":"AlphaAlign: Incentivizing Safety Alignment with Extremely Simplified Reinforcement Learning","created":"2025-07-20","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":"2410.10862","title":"Superficial Safety Alignment Hypothesis","created":"2024-10-07","cited_by_count":14,"area":"alignment"},{"id":"2502.01208","title":"On Almost Surely Safe Alignment of Large Language Models at Inference-Time","created":"2025-02-03","cited_by_count":14,"area":"alignment"},{"id":"2407.16637","title":"Course-Correction: Safety Alignment Using Synthetic Preferences","created":"2024-07-23","cited_by_count":13,"area":"alignment"},{"id":"2507.21182","title":"SDD: Self-Degraded Defense against Malicious Fine-tuning","created":"2025-07-27","cited_by_count":13,"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":"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":"2512.11391","title":"Mitigating the Safety Alignment Tax with Null-Space Constrained Policy Optimization","created":"2025-12-12","cited_by_count":11,"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":"2510.02833","title":"Attack via Overfitting: 10-shot Benign Fine-tuning to Jailbreak LLMs","created":"2025-10-03","cited_by_count":10,"area":"alignment"},{"id":"2502.16825","title":"Finding the Sweet Spot: Preference Data Construction for Scaling Preference Optimization","created":"2025-02-24","cited_by_count":9,"area":"alignment"},{"id":"2505.17072","title":"Safety Alignment Can Be Not Superficial With Explicit Safety Signals","created":"2025-05-19","cited_by_count":9,"area":"alignment"},{"id":"2506.06636","title":"SafeLawBench: Towards Safe Alignment of Large Language Models","created":"2025-06-07","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":"2512.13655","title":"Comparative Analysis of LLM Abliteration Methods: A Cross-Architecture Evaluation","created":"2025-12-15","cited_by_count":9,"area":"alignment"},{"id":"2602.16660","title":"Align Once, Benefit Multilingually: Enforcing Multilingual Consistency for LLM Safety Alignment","created":"2026-02-18","cited_by_count":9,"area":"alignment"},{"id":"2601.01887","title":"Safety at One Shot: Patching Fine-Tuned LLMs with A Single Instance","created":"2026-01-05","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":"2503.19041","title":"LookAhead Tuning: Safer Language Models via Partial Answer Previews","created":"2025-03-24","cited_by_count":7,"area":"alignment"},{"id":"2505.20081","title":"Inference-time Alignment in Continuous Space","created":"2025-05-26","cited_by_count":7,"area":"alignment"},{"id":"2603.11388","title":"Deactivating Refusal Triggers: Understanding and Mitigating Overrefusal in Safety Alignment","created":"2026-03-12","cited_by_count":7,"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":"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":"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":"2510.10085","title":"Pharmacist: Safety Alignment Data Curation for Large Language Models against Harmful Fine-tuning","created":"2025-10-11","cited_by_count":6,"area":"alignment"},{"id":"2601.18292","title":"TriPlay-RL: Tri-Role Self-Play Reinforcement Learning for LLM Safety Alignment","created":"2026-01-26","cited_by_count":6,"area":"alignment"},{"id":"2605.02946","title":"RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs","created":"2026-05-01","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":"2504.02193","title":"More is Less: The Pitfalls of Multi-Model Synthetic Preference Data in DPO Safety Alignment","created":"2025-04-03","cited_by_count":5,"area":"alignment"},{"id":"2505.17122","title":"Shallow Preference Signals: Large Language Model Aligns Even Better with Truncated Data?","created":"2025-05-21","cited_by_count":5,"area":"alignment"},{"id":"2506.04679","title":"Normative Conflicts and Shallow AI Alignment","created":"2025-06-05","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":"2511.18039","title":"Curvature-Aware Safety Restoration In LLMs Fine-Tuning","created":"2025-11-22","cited_by_count":5,"area":"alignment"},{"id":"2503.10093","title":"Efficient Safety Alignment of Large Language Models via Preference Re-ranking and Representation-based Reward Modeling","created":"2025-03-13","cited_by_count":4,"area":"alignment"},{"id":"2505.11861","title":"Fair-PP: A Synthetic Dataset for Aligning LLM with Personalized Preferences of Social Equity","created":"2025-05-17","cited_by_count":4,"area":"alignment"},{"id":"2505.14607","title":"sudoLLM: On Multi-role Alignment of Language Models","created":"2025-05-20","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":"2511.06778","title":"SAFENLIDB: A Privacy-Preserving Safety Alignment Framework for LLM-based Natural Language Database Interfaces","created":"2025-11-10","cited_by_count":4,"area":"alignment"},{"id":"2512.09403","title":"Black-Box Behavioral Distillation Breaks Safety Alignment in Medical LLMs","created":"2025-12-10","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":"2602.07340","title":"Revisiting Robustness for LLM Safety Alignment via Selective Geometry Control","created":"2026-02-07","cited_by_count":4,"area":"alignment"},{"id":"2603.07445","title":"Few Tokens, Big Leverage: Preserving Safety Alignment by Constraining Safety Tokens during Fine-tuning","created":"2026-03-08","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":"2606.24014","title":"Reinforcement Learning Towards Broadly and Persistently Beneficial Models","created":"2026-06-22","cited_by_count":4,"area":"alignment"},{"id":"2502.02153","title":"Vulnerability Mitigation for Safety-Aligned Language Models via Debiasing","created":"2025-02-04","cited_by_count":3,"area":"alignment"},{"id":"2505.11790","title":"JULI: Jailbreak Large Language Models by Self-Introspection","created":"2025-05-17","cited_by_count":3,"area":"alignment"},{"id":"2506.04462","title":"Watermarking Degrades Alignment in Language Models: Analysis and Mitigation","created":"2025-06-04","cited_by_count":3,"area":"alignment"},{"id":"2506.07402","title":"Beyond Jailbreaks: Revealing Stealthier and Broader LLM Security Risks Stemming from Alignment Failures","created":"2025-06-09","cited_by_count":3,"area":"alignment"},{"id":"2507.16679","title":"PICACO: Pluralistic In-Context Value Alignment of LLMs via Total Correlation Optimization","created":"2025-07-22","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":"2601.20838","title":"Reward Models Inherit Value Biases from Pretraining","created":"2026-01-28","cited_by_count":3,"area":"alignment"},{"id":"2602.04448","title":"RASA: Routing-Aware Safety Alignment for Mixture-of-Experts Models","created":"2026-02-04","cited_by_count":3,"area":"alignment"},{"id":"2602.08813","title":"Robust Policy Optimization to Prevent Catastrophic Forgetting","created":"2026-02-09","cited_by_count":3,"area":"alignment"},{"id":"2605.02971","title":"Multilingual Safety Alignment via Self-Distillation","created":"2026-05-03","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":"2502.11681","title":"RIDE: Enhancing Large Language Model Alignment through Restyled In-Context Learning Demonstration Exemplars","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":"2511.06512","title":"EASE: Practical and Efficient Safety Alignment for Small Language Models","created":"2025-11-09","cited_by_count":2,"area":"alignment"},{"id":"2601.07200","title":"Safeguarding LLM Fine-tuning via Push-Pull Distributional Alignment","created":"2026-01-12","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":"2603.03081","title":"TAO-Attack: Toward Advanced Optimization-Based Jailbreak Attacks for Large Language Models","created":"2026-03-03","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":"2603.10243","title":"GR-SAP: Generative Replay for Safety Alignment Preservation during Fine-Tuning","created":"2026-03-10","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":"2604.02174","title":"Quantifying Self-Preservation Bias in Large Language Models","created":"2026-04-02","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":"2606.21906","title":"Deeper is Not Always Better: Mitigating the Alignment Tax via Confident Layer Decoding","created":"2026-06-20","cited_by_count":2,"area":"alignment"},{"id":"2502.00669","title":"Safety Alignment Depth in Large Language Models: A Markov Chain Perspective","created":"2025-02-02","cited_by_count":1,"area":"alignment"},{"id":"2505.19700","title":"Leveraging Importance Sampling to Detach Alignment Modules from Large Language Models","created":"2025-05-26","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":"2506.06444","title":"Saffron-1: Safety Inference Scaling","created":"2025-06-06","cited_by_count":1,"area":"alignment"},{"id":"2506.13285","title":"DualEdit: Mitigating Safety Fallback in LLM Backdoor Editing via Affirmation-Refusal Regulation","created":"2025-06-16","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":"2510.02493","title":"Beyond Imitation: Recovering Dense Rewards from Demonstrations","created":"2025-10-02","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":"2603.16210","title":"MOSAIC: Composable Safety Alignment with Modular Control Tokens","created":"2026-03-17","cited_by_count":1,"area":"alignment"},{"id":"2603.22829","title":"Improving Safety Alignment via Balanced Direct Preference Optimization","created":"2026-03-24","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":"2605.29396","title":"Aligned but Fragile: Enhancing LLM Safety Robustness via Zeroth-Order Optimization","created":"2026-05-28","cited_by_count":1,"area":"alignment"},{"id":"2606.09735","title":"The Neutral Mask: How Alignment Training Provides Shallow Alignment while Leaving Partisan Structure Intact in a Large Language Model","created":"2026-06-08","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":"2607.02714","title":"Not All Refusals Are Equal: How Safety Alignment Fails Cybersecurity at Scale","created":"2026-07-02","cited_by_count":1,"area":"alignment"}],"next_cursor":"100"}