{"total":67,"rows":[{"id":"2507.11878","title":"LLMs Encode Harmfulness and Refusal Separately","created":"2025-07-16","cited_by_count":88,"area":"safety"},{"id":"2502.07557","title":"JBShield: Defending Large Language Models from Jailbreak Attacks through Activated Concept Analysis and Manipulation","created":"2025-02-11","cited_by_count":54,"area":"safety"},{"id":"2506.07022","title":"AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraint","created":"2025-06-08","cited_by_count":49,"area":"safety"},{"id":"2504.14492","title":"FairSteer: Inference Time Debiasing for LLMs with Dynamic Activation Steering","created":"2025-04-20","cited_by_count":40,"area":"mechinterp"},{"id":"2505.17306","title":"Refusal Direction is Universal Across Safety-Aligned Languages","created":"2025-05-22","cited_by_count":38,"area":null},{"id":"2505.20977","title":"Evaluating and Steering Modality Preferences in Multimodal Large Language Model","created":"2025-05-27","cited_by_count":33,"area":"mllm"},{"id":"2504.01550","title":"Representation Bending for Large Language Model Safety","created":"2025-04-02","cited_by_count":30,"area":"safety"},{"id":"2509.22067","title":"The Rogue Scalpel: Activation Steering Compromises LLM Safety","created":"2025-09-26","cited_by_count":21,"area":"safety"},{"id":"2602.17560","title":"ODESteer: A Unified ODE-Based Steering Framework for LLM Alignment","created":"2026-02-19","cited_by_count":19,"area":"alignment"},{"id":"2506.00253","title":"Aligned but Blind: Alignment Increases Implicit Bias by Reducing Awareness of Race","created":"2025-05-30","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":"2511.01815","title":"KV Cache Transform Coding for Compact Storage in LLM Inference","created":"2025-11-03","cited_by_count":13,"area":"inference"},{"id":"2602.22983","title":"Obscure but Effective: Classical Chinese Jailbreak Prompt Optimization via Bio-Inspired Search","created":"2026-02-26","cited_by_count":13,"area":"safety"},{"id":"2502.16903","title":"GuidedBench: Measuring and Mitigating the Evaluation Discrepancies of In-the-wild LLM Jailbreak Methods","created":"2025-02-24","cited_by_count":12,"area":"safety"},{"id":"2512.12069","title":"Rethinking Jailbreak Detection of Large Vision Language Models with Representational Contrastive Scoring","created":"2025-12-12","cited_by_count":11,"area":"lr"},{"id":"2502.18156","title":"Can LLMs Explain Themselves Counterfactually?","created":"2025-02-25","cited_by_count":10,"area":null},{"id":"2505.19610","title":"JailBound: Jailbreaking Internal Safety Boundaries of Vision-Language Models","created":"2025-05-26","cited_by_count":10,"area":"mllm"},{"id":"2506.05451","title":"Interpretation Meets Safety: A Survey on Interpretation Methods and Tools for Improving LLM Safety","created":"2025-06-05","cited_by_count":10,"area":"safety"},{"id":"2504.10227","title":"Probing then Editing Response Personality of Large Language Models","created":"2025-04-14","cited_by_count":9,"area":null},{"id":"2511.08525","title":"Investigating CoT Monitorability in Large Reasoning Models","created":"2025-11-11","cited_by_count":9,"area":"lr"},{"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":"2510.05969","title":"Probing the Difficulty Perception Mechanism of Large Language Models","created":"2025-10-07","cited_by_count":8,"area":null},{"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":"2509.01631","title":"Unraveling LLM Jailbreaks Through Safety Knowledge Neurons","created":"2025-09-01","cited_by_count":7,"area":"safety"},{"id":"2512.05485","title":"TeleAI-Safety: A comprehensive LLM jailbreaking benchmark towards attacks, defenses, and evaluations","created":"2025-12-05","cited_by_count":7,"area":"safety"},{"id":"2601.17887","title":"When Personalization Legitimizes Risks: Uncovering Safety Vulnerabilities in Personalized Dialogue Agents","created":"2026-01-25","cited_by_count":7,"area":"agents"},{"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":"2505.15524","title":"Evaluate Bias without Manual Test Sets: A Concept Representation Perspective for LLMs","created":"2025-05-21","cited_by_count":6,"area":"mechinterp"},{"id":"2509.18127","title":"Safe-SAIL: Towards a Fine-grained Safety Landscape of Large Language Models via Sparse Autoencoder Interpretation Framework","created":"2025-09-11","cited_by_count":6,"area":"mechinterp"},{"id":"2503.11185","title":"Bleeding Pathways: Vanishing Discriminability in LLM Hidden States Fuels Jailbreak Attacks","created":"2025-03-14","cited_by_count":5,"area":"safety"},{"id":"2504.11337","title":"REWARD CONSISTENCY: Improving Multi-Objective Alignment from a Data-Centric Perspective","created":"2025-04-15","cited_by_count":5,"area":"alignment"},{"id":"2507.06043","title":"CAVGAN: Unifying Jailbreak and Defense of LLMs via Generative Adversarial Attacks on their Internal Representations","created":"2025-07-08","cited_by_count":5,"area":"safety"},{"id":"2509.10931","title":"Harmful Prompt Laundering: Jailbreaking LLMs with Abductive Styles and Symbolic Encoding","created":"2025-09-13","cited_by_count":5,"area":"safety"},{"id":"2603.13249","title":"Steering at the Source: Style Modulation Heads for Robust Persona Control","created":"2026-02-24","cited_by_count":5,"area":"mechinterp"},{"id":"2505.13862","title":"PandaGuard: Systematic Evaluation of LLM Safety against Jailbreaking Attacks","created":"2025-05-20","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":"2509.25271","title":"RADAR: A Risk-Aware Dynamic Multi-Agent Framework for LLM Safety Evaluation via Role-Specialized Collaboration","created":"2025-09-28","cited_by_count":4,"area":"agents"},{"id":"2502.02153","title":"Vulnerability Mitigation for Safety-Aligned Language Models via Debiasing","created":"2025-02-04","cited_by_count":3,"area":"alignment"},{"id":"2412.20087","title":"On the Validity of Traditional Vulnerability Scoring Systems for Adversarial Attacks against LLMs","created":"2024-12-28","cited_by_count":2,"area":null},{"id":"2504.06659","title":"Leveraging Machine Unlearning for Cost-Efficient Preference Alignment","created":"2025-04-09","cited_by_count":2,"area":"alignment"},{"id":"2506.07406","title":"InverseScope: Scalable Activation Inversion for Interpreting Large Language Models","created":"2025-06-09","cited_by_count":2,"area":"mechinterp"},{"id":"2509.11141","title":"When Smiley Turns Hostile: Interpreting How Emojis Trigger LLMs' Toxicity","created":"2025-09-14","cited_by_count":2,"area":null},{"id":"2509.23159","title":"ProtoTS: Learning Hierarchical Prototypes for Explainable Time Series Forecasting","created":"2025-09-27","cited_by_count":2,"area":null},{"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":"2603.20206","title":"Enhancing Safety of Large Language Models via Embedding Space Separation","created":"2026-03-01","cited_by_count":2,"area":"safety"},{"id":"2508.00555","title":"Activation-Guided Local Editing for Jailbreaking Attacks","created":"2025-08-01","cited_by_count":1,"area":"safety"},{"id":"2508.10404","title":"Layer-Wise Perturbations via Sparse Autoencoders for Adversarial Text Generation","created":"2025-08-14","cited_by_count":1,"area":"safety"},{"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":"2510.15430","title":"Learning to Detect Unknown Jailbreak Attacks in Large Vision-Language Models","created":"2025-10-17","cited_by_count":1,"area":"mllm"},{"id":"2512.06655","title":"Graph-Regularized Sparse Autoencoders for LLM Safety Steering","created":"2025-12-07","cited_by_count":1,"area":"safety"},{"id":"2512.14751","title":"One Leak Away: How Pretrained Model Exposure Amplifies Jailbreak Risks in Finetuned LLMs","created":"2025-12-14","cited_by_count":1,"area":"safety"},{"id":"2601.12460","title":"TrojanPraise: Jailbreak LLMs via Benign Fine-Tuning","created":"2026-01-18","cited_by_count":1,"area":"safety"},{"id":"2602.00158","title":"RAPTOR: Ridge-Adaptive Logistic Probes","created":"2026-01-29","cited_by_count":1,"area":null},{"id":"2603.05772","title":"Depth Charge: Jailbreak Large Language Models from Deep Safety Attention Heads","created":"2026-03-06","cited_by_count":1,"area":"safety"},{"id":"2603.13847","title":"Sirens' Whisper: Inaudible Near-Ultrasonic Jailbreaks of Speech-Driven LLMs","created":"2026-03-14","cited_by_count":1,"area":"safety"},{"id":"2605.27932","title":"When Think-with-Image Meets Safety: What Determines Multimodal Jailbreak Robustness?","created":"2026-05-27","cited_by_count":1,"area":"lr"},{"id":"2607.06807","title":"When Agents Go Rogue: Activation-Based Detection of Malicious Behaviors in Multi-Agent Systems","created":"2026-07-07","cited_by_count":1,"area":"agents"},{"id":"2506.04774","title":"Fine-Grained Interpretation of Political Opinions in Large Language Models","created":"2025-06-05","cited_by_count":0,"area":"mechinterp"},{"id":"2601.22169","title":"In Vino Veritas and Vulnerabilities: Examining LLM Safety via Drunk Language Inducement","created":"2026-01-19","cited_by_count":0,"area":"safety"},{"id":"2602.02536","title":"From Sparse Decisions to Dense Reasoning: A Multi-attribute Trajectory Paradigm for Multimodal Moderation","created":"2026-01-28","cited_by_count":0,"area":"safety"},{"id":"2604.11663","title":"Why Do Large Language Models Generate Harmful Content?","created":"2026-04-13","cited_by_count":0,"area":"safety"},{"id":"2604.17210","title":"Guardrails in Logit Space: Safety Token Regularization for LLM Alignment","created":"2026-04-19","cited_by_count":0,"area":"alignment"},{"id":"2605.20286","title":"Adaptive Probe-based Steering for Robust LLM Jailbreaking","created":"2026-05-19","cited_by_count":0,"area":"safety"},{"id":"2606.19755","title":"SafeSpec: Fast and Safe LLM via Dynamic Reflective Sampling","created":"2026-06-18","cited_by_count":0,"area":"inference"},{"id":"2608.26506","title":"A Single Suffix to Break Them All: Basin-Aware Jailbreaks for Merged Model Families","created":"2026-08-27","cited_by_count":0,"area":"safety"},{"id":"2609.09798","title":"CS-Guard: Benchmarking LLM Guardrails for Code Generation Security","created":"2026-09-09","cited_by_count":0,"area":"safety"},{"id":"2610.06174","title":"Anosognosia in LLMs: Probing Self-Awareness of Quantized Computational Substrate","created":"2026-10-05","cited_by_count":0,"area":null}],"next_cursor":null}