{"total":72,"rows":[{"id":"2507.11878","title":"LLMs Encode Harmfulness and Refusal Separately","created":"2025-07-16","cited_by_count":88,"area":"safety"},{"id":"2502.09674","title":"The Hidden Dimensions of LLM Alignment: A Multi-Dimensional Analysis of Orthogonal Safety Directions","created":"2025-02-13","cited_by_count":41,"area":"lr"},{"id":"2506.14697","title":"AGENTSAFE: Benchmarking the Safety of Embodied Agents on Hazardous Instructions","created":"2025-06-17","cited_by_count":38,"area":"mllm"},{"id":"2502.05759","title":"Reinforced Lifelong Editing for Language Models","created":"2025-02-09","cited_by_count":34,"area":null},{"id":"2601.14004","title":"Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models","created":"2026-01-20","cited_by_count":27,"area":"mechinterp"},{"id":"2506.17368","title":"SAFEx: Analyzing Vulnerabilities of MoE-Based LLMs via Stable Safety-critical Expert Identification","created":"2025-06-20","cited_by_count":18,"area":"alignment"},{"id":"2502.13095","title":"Understanding and Rectifying Safety Perception Distortion in VLMs","created":"2025-02-18","cited_by_count":17,"area":"mllm"},{"id":"2503.22048","title":"ThinkEdit: Interpretable Weight Editing to Mitigate Overly Short Thinking in Reasoning Models","created":"2025-03-27","cited_by_count":17,"area":"lr"},{"id":"2510.26418","title":"Chain-of-Thought Hijacking","created":"2025-10-30","cited_by_count":16,"area":"lr"},{"id":"2507.00841","title":"SafeMobile: Chain-level Jailbreak Detection and Automated Evaluation for Multimodal Mobile Agents","created":"2025-07-01","cited_by_count":15,"area":"safety"},{"id":"2506.12430","title":"Pushing the Limits of Safety: A Technical Report on the ATLAS Challenge 2025","created":"2025-06-14","cited_by_count":12,"area":"mllm"},{"id":"2510.06036","title":"Refusal Falls off a Cliff: How Safety Alignment Fails in Reasoning?","created":"2025-10-07","cited_by_count":11,"area":"lr"},{"id":"2602.16823","title":"Formal Mechanistic Interpretability: Automated Circuit Discovery with Provable Guarantees","created":"2026-02-18","cited_by_count":11,"area":"mechinterp"},{"id":"2503.06269","title":"Using Mechanistic Interpretability to Craft Adversarial Attacks against Large Language Models","created":"2025-03-08","cited_by_count":10,"area":"mechinterp"},{"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":"2509.21761","title":"Backdoor Attribution: Elucidating and Controlling Backdoor in Language Models","created":"2025-09-26","cited_by_count":10,"area":"safety"},{"id":"2601.21996","title":"Mechanistic Data Attribution: Tracing the Training Origins of Interpretable LLM Units","created":"2026-01-29","cited_by_count":10,"area":"mechinterp"},{"id":"2511.08525","title":"Investigating CoT Monitorability in Large Reasoning Models","created":"2025-11-11","cited_by_count":9,"area":"lr"},{"id":"2602.08621","title":"Sparse Models, Sparse Safety: Unsafe Routes in Mixture-of-Experts LLMs","created":"2026-02-09","cited_by_count":9,"area":"moe"},{"id":"2508.02175","title":"Hidden in the Noise: Unveiling Backdoors in Audio LLMs Alignment through Latent Acoustic Pattern Triggers","created":"2025-08-04","cited_by_count":8,"area":"safety"},{"id":"2603.01285","title":"Attention Smoothing Is All You Need For Unlearning","created":"2026-03-01","cited_by_count":7,"area":"safety"},{"id":"2604.13627","title":"(How) Learning Rates Regulate Catastrophic Overtraining","created":"2026-04-15","cited_by_count":6,"area":null},{"id":"2605.05668","title":"Large Vision-Language Models Get Lost in Attention","created":"2026-05-07","cited_by_count":5,"area":"mllm"},{"id":"2505.16104","title":"Hierarchical Safety Realignment: Lightweight Restoration of Safety in Pruned Large Vision-Language Models","created":"2025-05-22","cited_by_count":4,"area":"mllm"},{"id":"2602.24021","title":"Steering and Rectifying Latent Representation Manifolds in Frozen Multi-modal LLMs for Video Anomaly Detection","created":"2026-02-27","cited_by_count":4,"area":"mllm"},{"id":"2603.20662","title":"Attention in Space: Functional Roles of VLM Heads for Spatial Reasoning","created":"2026-03-21","cited_by_count":4,"area":"lr"},{"id":"2603.23268","title":"SafeSeek: Universal Attribution of Safety Circuits in Language Models","created":"2026-03-24","cited_by_count":4,"area":"safety"},{"id":"2606.02530","title":"SafeSteer: Localized On-Policy Distillation for Efficient Safety Alignment","created":"2026-06-01","cited_by_count":4,"area":"alignment"},{"id":"2501.16534","title":"Targeting Alignment: Extracting Safety Classifiers of Aligned LLMs","created":"2025-01-27","cited_by_count":3,"area":"safety"},{"id":"2510.18470","title":"CircuitSeer: Mining High-Quality Data by Probing Mathematical Reasoning Circuits in LLMs","created":"2025-10-21","cited_by_count":3,"area":"lr"},{"id":"2602.13547","title":"AISA: Awakening Intrinsic Safety Awareness in Large Language Models against Jailbreak Attacks","created":"2026-02-14","cited_by_count":3,"area":"safety"},{"id":"2505.18799","title":"ALPS: Attention Localization and Pruning Strategy for Efficient Alignment of Large Language Models","created":"2025-05-24","cited_by_count":2,"area":"alignment"},{"id":"2511.07498","title":"Focusing on Language: Revealing and Exploiting Language Attention Heads in Multilingual Large Language Models","created":"2025-11-10","cited_by_count":2,"area":null},{"id":"2511.14182","title":"WebRec: Enhancing LLM-based Recommendations with Attention-guided RAG from Web","created":"2025-11-18","cited_by_count":2,"area":"rag"},{"id":"2512.10300","title":"Investigating The Functional Roles of Attention Heads in Vision Language Models: Evidence for Reasoning Modules","created":"2025-12-11","cited_by_count":2,"area":"lr"},{"id":"2602.03402","title":"Risk Awareness Injection: Calibrating Vision-Language Models for Safety without Compromising Utility","created":"2026-02-03","cited_by_count":2,"area":"lr"},{"id":"2602.09870","title":"Steer2Edit: From Activation Steering to Component-Level Editing","created":"2026-02-10","cited_by_count":2,"area":"mechinterp"},{"id":"2603.03335","title":"Compressed Sensing for Capability Localization in Large Language Models","created":"2026-02-11","cited_by_count":2,"area":"lr"},{"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":"2412.20476","title":"Cut the Deadwood Out: Backdoor Purification via Guided Module Substitution","created":"2024-12-29","cited_by_count":1,"area":null},{"id":"2509.25843","title":"ASGuard: Activation-Scaling Guard to Mitigate Targeted Jailbreaking Attack","created":"2025-09-30","cited_by_count":1,"area":"safety"},{"id":"2601.13631","title":"ContiguousKV: Accelerating LLM Prefill with Granularity-Aligned KV Cache Management","created":"2026-01-20","cited_by_count":1,"area":"inference"},{"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":"2604.11309","title":"The Salami Slicing Threat: Exploiting Cumulative Risks in LLM Systems","created":"2026-04-13","cited_by_count":1,"area":"safety"},{"id":"2604.15294","title":"How Do LLMs and VLMs Understand Viewpoint Rotation Without Vision? An Interpretability Study","created":"2026-04-16","cited_by_count":1,"area":null},{"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":"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"},{"id":"2609.25049","title":"Mitigating LLM Over-Refusal via Dynamic Semantic Routing Calibration","created":"2026-09-06","cited_by_count":1,"area":"safety"},{"id":"2510.15948","title":"VisuoAlign: Safety Alignment of LVLMs with Multimodal Tree Search","created":"2025-10-10","cited_by_count":0,"area":"lr"},{"id":"2512.00918","title":"Sparse Neuron Ablation Triggers Catastrophic Collapse of the Language Core in Large Vision-Language Models","created":"2025-11-30","cited_by_count":0,"area":"mllm"},{"id":"2512.05518","title":"Matching Ranks Over Probability Yields Truly Deep Safety Alignment","created":"2025-12-05","cited_by_count":0,"area":"alignment"},{"id":"2601.04262","title":"Safety-Utility Conflicts Are Not Global: Surgical Alignment via Head-Level Diagnosis","created":"2026-01-07","cited_by_count":0,"area":"alignment"},{"id":"2601.15801","title":"Attributing and Exploiting Safety Vectors through Global Optimization in Large Language Models","created":"2026-01-22","cited_by_count":0,"area":"safety"},{"id":"2602.04856","title":"CoT is Not the Chain of Truth: An Empirical Internal Analysis of Reasoning LLMs for Fake News Generation","created":"2026-02-04","cited_by_count":0,"area":"lr"},{"id":"2603.21317","title":"Stream separation improves Bregman conditioning in transformers","created":"2026-03-22","cited_by_count":0,"area":null},{"id":"2604.11663","title":"Why Do Large Language Models Generate Harmful Content?","created":"2026-04-13","cited_by_count":0,"area":"safety"},{"id":"2605.00123","title":"Minimal, Local, Causal Explanations for Jailbreak Success in Large Language Models","created":"2026-04-30","cited_by_count":0,"area":"safety"},{"id":"2605.17971","title":"Babel: Jailbreaking Safety Attention via Obfuscation Distribution Optimized Sampling","created":"2026-05-18","cited_by_count":0,"area":"alignment"},{"id":"2605.21558","title":"From Parameters to Data: A Task-Parameter-Guided Fine-Tuning Pipeline for Efficient LLM Alignment","created":"2026-05-20","cited_by_count":0,"area":"alignment"},{"id":"2605.27997","title":"Where Does Toxicity Live? Mechanistic Localization and Targeted Suppression in Language Models","created":"2026-05-27","cited_by_count":0,"area":"safety"},{"id":"2606.13209","title":"Understanding helpfulness and harmless tension in reward models","created":"2026-06-11","cited_by_count":0,"area":"alignment"},{"id":"2607.25873","title":"How Do LLMs Read Bug Reports? An Empirical Study of Attention in LLMs for Automated Program Repair","created":"2026-07-28","cited_by_count":0,"area":null},{"id":"2608.04322","title":"DataRx: Missingness-Aware Sampling for Safer Large Language Model Task-Specific Fine-Tuning","created":"2026-08-05","cited_by_count":0,"area":null},{"id":"2608.11138","title":"Attention-Path Fragility as an Uncertainty Signal in Large Language Models","created":"2026-08-11","cited_by_count":0,"area":null},{"id":"2608.26222","title":"NeuronFuzz: Safety Neuron Guided Fuzzing for LLM Safety Evaluation","created":"2026-08-26","cited_by_count":0,"area":"alignment"},{"id":"2608.27580","title":"LongGuard: Mechanistic Analysis and Training-Free Mitigation of Long-Context Failure in Safety Guardrails","created":"2026-08-27","cited_by_count":0,"area":"safety"},{"id":"2608.30403","title":"Why Are LLM Backdoor Defenses Fragmented? A Feature-Level Explanation with Sparse Autoencoders","created":"2026-08-31","cited_by_count":0,"area":"safety"},{"id":"2609.04721","title":"Locating and Steering Refusal Beyond Attention","created":"2026-09-04","cited_by_count":0,"area":null},{"id":"2609.23700","title":"When the Agent Becomes the Kernel: A Systematization of Security on the Path to AI-Native Operating Systems","created":"2026-09-20","cited_by_count":0,"area":"agents"},{"id":"2609.35117","title":"Tool Mediation Alters Refusal Mechanisms in Large Language Models","created":"2026-09-28","cited_by_count":0,"area":null},{"id":"2610.19452","title":"Are You Reading the Right Thing? Rethinking Internal Signals for Prompt-Injection Detection","created":"2026-10-14","cited_by_count":0,"area":"safety"}],"next_cursor":null}