{"total":53,"rows":[{"id":"2410.13708","title":"On the Role of Attention Heads in Large Language Model Safety","created":"2024-10-17","cited_by_count":72,"area":"safety"},{"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":"2411.11114","title":"JailbreakLens: Interpreting Jailbreak Mechanism in the Lens of Representation and Circuit","created":"2024-11-17","cited_by_count":22,"area":"safety"},{"id":"2506.07596","title":"TwinBreak: Jailbreaking LLM Security Alignments based on Twin Prompts","created":"2025-06-09","cited_by_count":19,"area":"safety"},{"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":"2505.15710","title":"Advancing LLM Safe Alignment with Safety Representation Ranking","created":"2025-05-21","cited_by_count":16,"area":"safety"},{"id":"2510.26418","title":"Chain-of-Thought Hijacking","created":"2025-10-30","cited_by_count":16,"area":"lr"},{"id":"2410.10862","title":"Superficial Safety Alignment Hypothesis","created":"2024-10-07","cited_by_count":14,"area":"alignment"},{"id":"2501.19180","title":"Enhancing Model Defense Against Jailbreaks with Proactive Safety Reasoning","created":"2025-01-31","cited_by_count":11,"area":"lr"},{"id":"2504.07440","title":"Model Utility Law: Evaluating LLMs beyond Performance through Mechanism Interpretable Metric","created":"2025-04-10","cited_by_count":11,"area":null},{"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":"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":"2503.23306","title":"Focus Directions Make Your Language Models Pay More Attention to Relevant Contexts","created":"2025-03-30","cited_by_count":9,"area":null},{"id":"2408.15207","title":"Understanding the Effectiveness of Coverage Criteria for Large Language Models: A Special Angle from Jailbreak Attacks","created":"2024-08-27","cited_by_count":8,"area":"safety"},{"id":"2510.01088","title":"Safety Instincts: LLMs Learn to Trust Their Internal Compass for Self-Defense","created":"2025-10-01","cited_by_count":4,"area":"safety"},{"id":"2511.00382","title":"Efficiency vs. Alignment: Investigating Safety and Fairness Risks in Parameter-Efficient Fine-Tuning of LLMs","created":"2025-11-01","cited_by_count":4,"area":null},{"id":"2602.08741","title":"Large Language Lobotomy: Jailbreaking Mixture-of-Experts via Expert Silencing","created":"2026-02-09","cited_by_count":4,"area":"safety"},{"id":"2603.23268","title":"SafeSeek: Universal Attribution of Safety Circuits in Language Models","created":"2026-03-24","cited_by_count":4,"area":"safety"},{"id":"2605.08513","title":"A Single Neuron Is Sufficient to Bypass Safety Alignment in Large Language Models","created":"2026-05-08","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":"2602.21346","title":"Alignment-Weighted DPO: A principled reasoning approach to improve safety alignment","created":"2026-02-24","cited_by_count":2,"area":"lr"},{"id":"2603.03335","title":"Compressed Sensing for Capability Localization in Large Language Models","created":"2026-02-11","cited_by_count":2,"area":"lr"},{"id":"2608.08029","title":"Do All LLMs Know When They're Being Harmful? A Reproducibility Study of Latent-Space Safety Probes Across Model Families","created":"2026-08-08","cited_by_count":2,"area":"safety"},{"id":"2410.18114","title":"Bridging Today and the Future of Humanity: AI Safety in 2024 and Beyond","created":"2024-10-09","cited_by_count":1,"area":null},{"id":"2509.03985","title":"NeuroBreak: Unveil Internal Jailbreak Mechanisms in Large Language Models","created":"2025-09-04","cited_by_count":1,"area":"alignment"},{"id":"2511.08484","title":"Patching LLM Like Software: A Lightweight Method for Improving Safety Policy in Large Language Models","created":"2025-11-11","cited_by_count":1,"area":null},{"id":"2602.04224","title":"RAPO: Risk-Aware Preference Optimization for Generalizable Safe Reasoning","created":"2026-02-04","cited_by_count":1,"area":"lr"},{"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":"2604.16889","title":"Prune, Interpret, Evaluate: A Cross-Layer Transcoder-Native Framework for Efficient Circuit Discovery via Feature Attribution","created":"2026-04-18","cited_by_count":1,"area":"mechinterp"},{"id":"2608.23264","title":"Hidden in the Request: Explaining Unethical LLM Compliance through Token Relevance","created":"2026-08-24","cited_by_count":1,"area":null},{"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":"2609.16204","title":"Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration","created":"2026-09-14","cited_by_count":1,"area":"safety"},{"id":"2609.19472","title":"Safety Beyond the Interface: Detecting Harm via Latent States in Large Language Models","created":"2026-09-16","cited_by_count":1,"area":"safety"},{"id":"2504.08000","title":"Neuron-level Balance between Stability and Plasticity in Deep Reinforcement Learning","created":"2025-04-09","cited_by_count":0,"area":"deeprl"},{"id":"2605.27836","title":"Symmetry Defeats Auditing","created":"2026-05-27","cited_by_count":0,"area":null},{"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":"2606.19831","title":"Leverage Is Not Reach: A Control-Window Law for Single-Neuron Steering in Language Models","created":"2026-06-18","cited_by_count":0,"area":"alignment"},{"id":"2607.05356","title":"ReCal3R: Reliability-Calibrated Learning Rates for Streaming 3D Reconstruction","created":"2026-07-06","cited_by_count":0,"area":null},{"id":"2607.28959","title":"Efficient LLM Adversarial Training via Low-Rank Defense and Circuit-Guided Surrogates","created":"2026-07-31","cited_by_count":0,"area":null},{"id":"2608.22490","title":"Who Pays More for Safety? Measuring the Disparate Cost of Safety Alignment across Languages","created":"2026-08-23","cited_by_count":0,"area":"alignment"},{"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":"2608.30750","title":"Do VLMs Share Safety Neurons Across Modalities?","created":"2026-08-31","cited_by_count":0,"area":"mllm"},{"id":"2609.02293","title":"SEAL: Reinforcing Global Safety in Mixture-of-Experts through Shared Expert ALignment","created":"2026-09-02","cited_by_count":0,"area":"moe"},{"id":"2609.18471","title":"First Token Matters: Understanding Safety Collapse in Large Reasoning Models","created":"2026-09-16","cited_by_count":0,"area":"lr"},{"id":"2609.22144","title":"Multilingual Safety Signals Are Multi-Layered: Filtering Safety-Degrading Data for Safer LLMs","created":"2026-08-26","cited_by_count":0,"area":"alignment"},{"id":"2609.32434","title":"From Latents to Wires: Surgical Post-Editing on Large Language Models","created":"2026-09-26","cited_by_count":0,"area":"safety"},{"id":"2609.34771","title":"When Do Model Internals Help? Exploring the Role of Representation Engineering in LLM Safety","created":"2026-09-28","cited_by_count":0,"area":"alignment"},{"id":"2609.35117","title":"Tool Mediation Alters Refusal Mechanisms in Large Language Models","created":"2026-09-28","cited_by_count":0,"area":null},{"id":"2610.04316","title":"From Latent Space to Jacobian Space: Measuring, Evading, and Training Against Safety-Content Accessibility","created":"2026-10-03","cited_by_count":0,"area":null},{"id":"2610.05541","title":"Don't Judge an LLM Only by Its Activations: Discovering Suppressed Safety Features via Counterfactual Activation Potential","created":"2026-10-04","cited_by_count":0,"area":"safety"},{"id":"2610.09600","title":"SafeEvo: Deciphering the Safety Alignment Mechanism and Evolution in Language Models","created":"2026-10-07","cited_by_count":0,"area":"alignment"}],"next_cursor":null}