{"total":115,"rows":[{"id":"2507.11878","title":"LLMs Encode Harmfulness and Refusal Separately","created":"2025-07-16","cited_by_count":88,"area":"safety"},{"id":"2506.11618","title":"Convergent Linear Representations of Emergent Misalignment","created":"2025-06-13","cited_by_count":71,"area":null},{"id":"2506.07022","title":"AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraint","created":"2025-06-08","cited_by_count":49,"area":"safety"},{"id":"2601.04480","title":"When Models Manipulate Manifolds: The Geometry of a Counting Task","created":"2026-01-08","cited_by_count":46,"area":null},{"id":"2604.28119","title":"Do Sparse Autoencoders Capture Concept Manifolds?","created":"2026-04-30","cited_by_count":31,"area":"mechinterp"},{"id":"2602.02132","title":"There Is More to Refusal in Large Language Models than a Single Direction","created":"2026-02-02","cited_by_count":26,"area":"mechinterp"},{"id":"2601.16644","title":"Sycophancy Hides Linearly in the Attention Heads","created":"2026-01-23","cited_by_count":23,"area":null},{"id":"2509.09708","title":"Beyond I'm Sorry, I Can't: Dissecting Large Language Model Refusal","created":"2025-09-07","cited_by_count":19,"area":"safety"},{"id":"2506.08390","title":"On Reasoning Strength Planning in Large Reasoning Models","created":"2025-06-10","cited_by_count":13,"area":"lr"},{"id":"2509.13281","title":"RepIt: Steering Language Models with Concept-Specific Refusal Vectors","created":"2025-09-16","cited_by_count":12,"area":null},{"id":"2602.15293","title":"The Information Geometry of Softmax: Probing and Steering","created":"2026-02-17","cited_by_count":11,"area":null},{"id":"2603.09313","title":"Curveball Steering: The Right Direction To Steer Isn't Always Linear","created":"2026-03-10","cited_by_count":11,"area":"mechinterp"},{"id":"2510.04217","title":"MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steering","created":"2025-10-05","cited_by_count":10,"area":"mllm"},{"id":"2603.24543","title":"Analysing the Safety Pitfalls of Steering Vectors","created":"2026-03-25","cited_by_count":10,"area":"safety"},{"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":"2602.02343","title":"Why Steering Works: Toward a Unified View of Language Model Parameter Dynamics","created":"2026-02-02","cited_by_count":9,"area":null},{"id":"2508.01916","title":"Decomposing Representation Space into Interpretable Subspaces with Unsupervised Learning","created":"2025-08-03","cited_by_count":8,"area":"mechinterp"},{"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":"2507.21141","title":"The Geometry of Harmfulness in LLMs through Subconcept Probing","created":"2025-07-23","cited_by_count":7,"area":null},{"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":"2511.04316","title":"AdversariaLLM: A Unified and Modular Toolbox for LLM Robustness Research","created":"2025-11-06","cited_by_count":6,"area":"safety"},{"id":"2506.15751","title":"Sysformer: Safeguarding Frozen Large Language Models with Adaptive System Prompts","created":"2025-06-18","cited_by_count":5,"area":"safety"},{"id":"2505.21800","title":"From Directions to Cones: Exploring Multidimensional Representations of Propositional Facts in LLMs","created":"2025-05-27","cited_by_count":4,"area":null},{"id":"2506.20666","title":"Cognitive models can reveal interpretable value trade-offs in language models","created":"2025-06-25","cited_by_count":3,"area":null},{"id":"2507.03167","title":"Where Do Reasoning Models Refuse?","created":"2025-07-03","cited_by_count":3,"area":"lr"},{"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":"2603.03163","title":"Conditioned Activation Transport for T2I Safety Steering","created":"2026-03-03","cited_by_count":3,"area":"mechinterp"},{"id":"2603.22061","title":"On the Failure of Topic-Matched Contrast Baselines in Multi-Directional Refusal Abliteration","created":"2026-03-23","cited_by_count":3,"area":"safety"},{"id":"2604.04385","title":"How Alignment Routes: Localizing, Scaling, and Controlling Policy Circuits in Language Models","created":"2026-04-06","cited_by_count":3,"area":"mechinterp"},{"id":"2605.12726","title":"Before the Last Token: Diagnosing Final-Token Safety Probe Failures","created":"2026-05-12","cited_by_count":3,"area":"safety"},{"id":"2607.21356","title":"Emergent Misalignment Recruits a Pre-existing Persona Subspace","created":"2026-07-23","cited_by_count":3,"area":"alignment"},{"id":"2602.16977","title":"Fail-Closed Alignment for Large Language Models","created":"2026-02-19","cited_by_count":2,"area":"alignment"},{"id":"2603.27518","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":"2604.16824","title":"SafeDream: Safety World Model for Proactive Early Jailbreak Detection","created":"2026-04-18","cited_by_count":2,"area":"wm"},{"id":"2604.27019","title":"Dynamic Adversarial Fine-Tuning Reorganizes Refusal Geometry","created":"2026-04-29","cited_by_count":2,"area":"alignment"},{"id":"2605.05715","title":"Decodable but Not Corrected by Fixed Residual-Stream Linear Steering: Evidence from Medical LLM Failure Regimes","created":"2026-05-07","cited_by_count":2,"area":null},{"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":"2607.05790","title":"Controlling Tool Use with Heading-Specific Activation Steering","created":"2026-07-07","cited_by_count":2,"area":"agents"},{"id":"2607.13075","title":"The Entanglement Wall: Activation-Space Probes as Risk Detectors, Not Context Adjudicators","created":"2026-07-12","cited_by_count":2,"area":"safety"},{"id":"2512.06655","title":"Graph-Regularized Sparse Autoencoders for LLM Safety Steering","created":"2025-12-07","cited_by_count":1,"area":"safety"},{"id":"2601.01584","title":"Steerability of Instrumental-Convergence Tendencies in LLMs","created":"2026-01-04","cited_by_count":1,"area":null},{"id":"2602.02600","title":"Step-Wise Refusal Dynamics in Autoregressive and Diffusion Language Models","created":"2026-02-01","cited_by_count":1,"area":"diffusion"},{"id":"2602.16980","title":"Discovering Universal Activation Directions for PII Leakage in Language Models","created":"2026-02-19","cited_by_count":1,"area":null},{"id":"2604.11120","title":"Persona Non Grata: Single-Method Safety Evaluation Is Incomplete for Persona-Imbued LLMs","created":"2026-04-13","cited_by_count":1,"area":null},{"id":"2605.01913","title":"RefusalGuard: Geometry-Preserving Fine-Tuning for Safety in LLMs","created":"2026-05-03","cited_by_count":1,"area":"alignment"},{"id":"2605.21706","title":"Latent-space Attacks for Refusal Evasion in Language Models","created":"2026-05-20","cited_by_count":1,"area":"alignment"},{"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":"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":"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"},{"id":"2609.03887","title":"Beyond Shallow Alignment: How Post-Training Methods Determine Refusal Circuits And Steering Robustness","created":"2026-09-03","cited_by_count":1,"area":"alignment"},{"id":"2609.05575","title":"Capsule Lens: Locating and Tracking Concept Geometry in Model Representations","created":"2026-09-04","cited_by_count":1,"area":"mechinterp"},{"id":"2609.16204","title":"Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration","created":"2026-09-14","cited_by_count":1,"area":"safety"},{"id":"2509.12221","title":"MEUV: Achieving Fine-Grained Capability Activation in Large Language Models via Mutually Exclusive Unlock Vectors","created":"2025-09-04","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":"2602.14529","title":"Disentangling Deception and Hallucination Failures in LLMs","created":"2026-02-16","cited_by_count":0,"area":null},{"id":"2602.17778","title":"Asking Forever: Universal Activations Behind Turn Amplification in Conversational LLMs","created":"2026-02-19","cited_by_count":0,"area":null},{"id":"2603.15647","title":"Steering Frozen LLMs: Adaptive Social Alignment via Online Prompt Routing","created":"2026-03-04","cited_by_count":0,"area":"alignment"},{"id":"2604.27169","title":"Semantic Structure of Feature Space in Large Language Models","created":"2026-04-29","cited_by_count":0,"area":null},{"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":"2605.02958","title":"Tracing the Dynamics of Refusal: Exploiting Latent Refusal Trajectories for Robust Jailbreak Detection","created":"2026-05-02","cited_by_count":0,"area":"safety"},{"id":"2605.14347","title":"Exemplar Partitioning for Mechanistic Interpretability","created":"2026-05-14","cited_by_count":0,"area":"safety"},{"id":"2605.14749","title":"Non-linear Interventions on Large Language Models","created":"2026-05-14","cited_by_count":0,"area":"mechinterp"},{"id":"2605.17173","title":"Why Do Safety Guardrails Degrade Across Languages?","created":"2026-05-16","cited_by_count":0,"area":"safety"},{"id":"2605.20262","title":"Residual Paving: Diagnosing the Routing Bottleneck in Selective Refusal Editing","created":"2026-05-18","cited_by_count":0,"area":null},{"id":"2605.24535","title":"Steering Beyond the Support: Adversarial Training on Unsupervised Jailbroken Activation Simulation","created":"2026-05-23","cited_by_count":0,"area":"safety"},{"id":"2605.24583","title":"Measuring Alignment-Induced Activation Shifts Correctly: A Template-Controlled Difference-in-Differences Protocol","created":"2026-05-23","cited_by_count":0,"area":null},{"id":"2606.00686","title":"Dialectics of Alignment: Harnessing Unsafe Knowledge for Dynamic Safety Routing","created":"2026-05-30","cited_by_count":0,"area":"moe"},{"id":"2606.04168","title":"When Autoregressive Consistency Hurts Safety Alignment","created":"2026-06-02","cited_by_count":0,"area":"alignment"},{"id":"2606.13709","title":"LoMC: Localized Multidirectional Correction for Refusal Suppression in Routed Foundation Models","created":"2026-06-10","cited_by_count":0,"area":"moe"},{"id":"2606.16349","title":"From Refusal Geometry to Safety Geometry: Harmfulness--Refusal Coupling under Dynamic Adversarial Fine-Tuning","created":"2026-06-15","cited_by_count":0,"area":"alignment"},{"id":"2606.26161","title":"Refusal Lives Downstream of Persona in Chat Models","created":"2026-06-24","cited_by_count":0,"area":null},{"id":"2606.26982","title":"Auditing Framing-Sensitive Behavioral Instability in Large Language Models for Mental Health Interactions","created":"2026-06-25","cited_by_count":0,"area":"mechinterp"},{"id":"2606.28843","title":"The Heterogeneous Safety Impacts of Benign Multilingual Fine-Tuning","created":"2026-06-27","cited_by_count":0,"area":null},{"id":"2607.00572","title":"HARC: Coupling Harmfulness and Refusal Directions for Robust Safety Alignment","created":"2026-07-01","cited_by_count":0,"area":"alignment"},{"id":"2607.01859","title":"Safety Targeted Embedding Exploit via Refinement","created":"2026-07-02","cited_by_count":0,"area":null},{"id":"2607.05356","title":"ReCal3R: Reliability-Calibrated Learning Rates for Streaming 3D Reconstruction","created":"2026-07-06","cited_by_count":0,"area":null},{"id":"2607.10112","title":"Minionese: Comprehensive Benchmark and Mechanistic Study of Multilingual LLM Safety","created":"2026-07-11","cited_by_count":0,"area":"alignment"},{"id":"2608.03745","title":"Risky Business: Measuring The Faithfulness-Safety Tension","created":"2026-08-04","cited_by_count":0,"area":"lr"},{"id":"2608.08383","title":"Safety Cost of Steering Vectors Is Separable and Reducible","created":"2026-08-09","cited_by_count":0,"area":"mechinterp"},{"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":"2608.15772","title":"Broken Symmetry in LLM Refusal: Answer Release Is More Local Than Refusal Restoration","created":"2026-08-16","cited_by_count":0,"area":null},{"id":"2608.17202","title":"Fool's Gold: Defensive Deception Against Safety-Removal Attacks on Open-Weight Models","created":"2026-08-17","cited_by_count":0,"area":"alignment"},{"id":"2608.17836","title":"Leveraging Association Context Retrieval in Knowledge Edit- ing to Build White-Box Attacks on LLMs","created":"2026-08-18","cited_by_count":0,"area":null},{"id":"2608.21928","title":"GuardianBench: A Same-Scene Instruction-Contrastive Benchmark for Latent Contextual Risk in Embodied AI","created":"2026-08-22","cited_by_count":0,"area":null},{"id":"2608.25390","title":"Refusal geometry reflects refusal training: diverse refusal prefixes can raise stable rank and weaken refusal vector ablation attacks","created":"2026-08-26","cited_by_count":0,"area":"safety"},{"id":"2608.29109","title":"Recognition-Refusal Misalignment in LLMs: Why Models Answer Structurally Unanswerable Questions","created":"2026-08-29","cited_by_count":0,"area":null},{"id":"2608.30585","title":"The Safety Relay in Roleplay Jailbreaks: A Component-Resolved Causal Analysis of Harm Recognition and Refusal","created":"2026-08-31","cited_by_count":0,"area":"safety"},{"id":"2608.30986","title":"Controlling Refusal Behavior of LLMs via Stiefel-Constrained Rotation Steering","created":"2026-08-31","cited_by_count":0,"area":"mechinterp"},{"id":"2609.03438","title":"Do GUI Agents Know When Not to Act? Enabling Conflict-Aware Termination for Multimodal GUI Agents","created":"2026-09-03","cited_by_count":0,"area":"agents"},{"id":"2609.05797","title":"Safety Monitors Mostly Catch What the Model Already Refuses","created":"2026-09-05","cited_by_count":0,"area":"safety"},{"id":"2609.06934","title":"The Geometry of Refusal: Why Post-Hoc Safety Is Fragile and Pretraining-Time Safety Persists","created":"2026-09-07","cited_by_count":0,"area":"alignment"},{"id":"2609.06951","title":"Steering Interference Reflects the Model's Defaults, Not the Behavior Directions","created":"2026-09-07","cited_by_count":0,"area":"mechinterp"},{"id":"2609.09793","title":"How Fragile Is Safety Alignment at Frontier Scale? A Single-Direction Attack on a 320B MoE","created":"2026-09-09","cited_by_count":0,"area":"alignment"},{"id":"2609.10142","title":"Active Adaptation, Not Static Defense: Temporal Dynamics of Preventative Steering in Adversarial Fine-Tuning","created":"2026-09-09","cited_by_count":0,"area":null},{"id":"2609.14861","title":"Semantic Fibers and Cross-Gram Interference: A Calculus of Safety Drift in Overcomplete Representations","created":"2026-09-14","cited_by_count":0,"area":"safety"},{"id":"2609.22153","title":"SafeTune: A Unified Faithful Library for Auditing and Repairing Safety Drift in Fine-Tuned LLMs","created":"2026-08-26","cited_by_count":0,"area":null},{"id":"2609.22224","title":"From Trait Vectors to Circuits: Tracing Refusal and Sycophancy Through Language Models","created":"2026-09-02","cited_by_count":0,"area":null},{"id":"2609.22260","title":"Causal Localization of the Refusal Direction in Audio Language Models","created":"2026-09-06","cited_by_count":0,"area":null},{"id":"2609.25021","title":"\"As a Language Model...\": Chat Template Switches LLM Self-Referential Voice and Activation Steering Reproduces It","created":"2026-08-09","cited_by_count":0,"area":"mechinterp"}],"next_cursor":"100"}