{"total":84,"rows":[{"id":"2406.11717","title":"Refusal in Language Models Is Mediated by a Single Direction","created":"2024-06-17","cited_by_count":1032,"area":"safety"},{"id":"2402.16835","title":"Eight Methods to Evaluate Robust Unlearning in LLMs","created":"2024-02-26","cited_by_count":163,"area":"safety"},{"id":"2408.15221","title":"LLM Defenses Are Not Robust to Multi-Turn Human Jailbreaks Yet","created":"2024-08-27","cited_by_count":151,"area":"safety"},{"id":"2403.05030","title":"Defending Against Unforeseen Failure Modes with Latent Adversarial Training","created":"2024-03-08","cited_by_count":88,"area":"safety"},{"id":"2507.11878","title":"LLMs Encode Harmfulness and Refusal Separately","created":"2025-07-16","cited_by_count":88,"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":"2406.15518","title":"Steering Without Side Effects: Improving Post-Deployment Control of Language Models","created":"2024-06-21","cited_by_count":46,"area":"safety"},{"id":"2502.07218","title":"LLM Unlearning via Neural Activation Redirection","created":"2025-02-11","cited_by_count":41,"area":"safety"},{"id":"2406.12094","title":"Who's asking? User personas and the mechanics of latent misalignment","created":"2024-06-17","cited_by_count":40,"area":"safety"},{"id":"2406.09289","title":"Understanding Jailbreak Success: A Study of Latent Space Dynamics in Large Language Models","created":"2024-06-13","cited_by_count":37,"area":"safety"},{"id":"2504.01550","title":"Representation Bending for Large Language Model Safety","created":"2025-04-02","cited_by_count":30,"area":"safety"},{"id":"2411.03343","title":"What Features in Prompts Jailbreak LLMs? Investigating the Mechanisms Behind Attacks","created":"2024-11-02","cited_by_count":29,"area":"safety"},{"id":"2411.02631","title":"Extracting Unlearned Information from LLMs with Activation Steering","created":"2024-11-04","cited_by_count":21,"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":"2510.14005","title":"PIShield: Detecting Prompt Injection Attacks via Intrinsic LLM Features","created":"2025-10-15","cited_by_count":20,"area":"safety"},{"id":"2506.07891","title":"Video Unlearning via Low-Rank Refusal Vector","created":"2025-06-09","cited_by_count":13,"area":"safety"},{"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":"2603.24543","title":"Analysing the Safety Pitfalls of Steering Vectors","created":"2026-03-25","cited_by_count":10,"area":"safety"},{"id":"2604.22167","title":"Estimating Tail Risks in Language Model Output Distributions","created":"2026-04-24","cited_by_count":9,"area":"safety"},{"id":"2510.17633","title":"SARSteer: Safeguarding Large Audio-Language Models via Safe-Ablated Refusal Steering","created":"2025-10-20","cited_by_count":8,"area":"safety"},{"id":"2512.16602","title":"Refusal Steering: Fine-grained Control over LLM Refusal Behaviour for Sensitive Topics","created":"2025-12-18","cited_by_count":6,"area":"safety"},{"id":"2602.12418","title":"Sparse Autoencoders are Capable LLM Jailbreak Mitigators","created":"2026-02-12","cited_by_count":6,"area":"safety"},{"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":"2502.15594","title":"SafeInt: Shielding Large Language Models from Jailbreak Attacks via Safety-Aware Representation Intervention","created":"2025-02-21","cited_by_count":4,"area":"safety"},{"id":"2504.09604","title":"Mitigating Many-Shot Jailbreaking","created":"2025-04-13","cited_by_count":4,"area":"safety"},{"id":"2508.12072","title":"Mitigating Jailbreaks with Intent-Aware LLMs","created":"2025-08-16","cited_by_count":4,"area":"safety"},{"id":"2604.08169","title":"Activation Steering for Aligned Open-ended Generation without Sacrificing Coherence","created":"2026-04-09","cited_by_count":4,"area":"safety"},{"id":"2604.14865","title":"Segment-Level Coherence for Robust Harmful Intent Probing in LLMs","created":"2026-04-16","cited_by_count":3,"area":"safety"},{"id":"2508.16325","title":"ConceptGuard: Neuro-Symbolic Safety Guardrails via Sparse Interpretable Jailbreak Concepts","created":"2025-08-22","cited_by_count":2,"area":"safety"},{"id":"2603.14278","title":"Activation Surgery: Jailbreaking White-box LLMs without Touching the Prompt","created":"2026-03-15","cited_by_count":2,"area":"safety"},{"id":"2605.05443","title":"SLAM: Structural Linguistic Activation Marking for Language Models","created":"2026-05-06","cited_by_count":2,"area":"safety"},{"id":"2605.12765","title":"Inference-Time Machine Unlearning via Gated Activation Redirection","created":"2026-05-12","cited_by_count":2,"area":"safety"},{"id":"2606.03291","title":"Multilingual Unlearning in LLMs: Transfer, Dynamics, and Reversibility","created":"2026-06-02","cited_by_count":2,"area":"safety"},{"id":"2606.19535","title":"FloatDoor: Platform-Triggered Backdoors in LLMs","created":"2026-06-17","cited_by_count":2,"area":"safety"},{"id":"2607.19806","title":"OPIUM: Mitigating Steering Externalities and Over-Refusal via Dual Objective Latent Optimization","created":"2026-07-22","cited_by_count":2,"area":"safety"},{"id":"2502.05945","title":"Head-Specific Intervention Can Induce Misaligned AI Coordination in Large Language Models","created":"2025-02-09","cited_by_count":1,"area":"safety"},{"id":"2511.07107","title":"MENTOR: A Metacognition-Driven Self-Evolution Framework for Uncovering and Mitigating Implicit Domain Risks in LLMs","created":"2025-11-10","cited_by_count":1,"area":"safety"},{"id":"2511.14195","title":"N-GLARE: An Non-Generative Latent Representation-Efficient LLM Safety Evaluator","created":"2025-11-18","cited_by_count":1,"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":"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":"2603.10806","title":"Backdoor Directions in Vision Transformers","created":"2026-03-11","cited_by_count":1,"area":"safety"},{"id":"2605.12290","title":"Targeted Neuron Modulation via Contrastive Pair Search","created":"2026-05-12","cited_by_count":1,"area":"safety"},{"id":"2605.26526","title":"Open-Weight LLM Fine-Tuning Defenses are Susceptible to Simple Attacks","created":"2026-05-26","cited_by_count":1,"area":"safety"},{"id":"2608.28233","title":"REINS: Refusal-Enhanced Inhibitory Steering with Sparse Autoencoder Features","created":"2026-08-28","cited_by_count":1,"area":"safety"},{"id":"2508.21448","title":"When Models Refuse: Political Steerability and Feature Richness as Measures of Ideological Depth","created":"2025-08-29","cited_by_count":0,"area":"safety"},{"id":"2509.21400","title":"SafeSteer: Adaptive Subspace Steering for Efficient Jailbreak Defense in Vision-Language Models","created":"2025-09-24","cited_by_count":0,"area":"safety"},{"id":"2510.12672","title":"Keep Calm and Avoid Harmful Content: Concept Alignment and Latent Manipulation Towards Safer Answers","created":"2025-10-14","cited_by_count":0,"area":"safety"},{"id":"2511.11667","title":"Beyond Superficial Forgetting: Thorough Unlearning through Knowledge Density Estimation and Block Re-insertion","created":"2025-11-11","cited_by_count":0,"area":"safety"},{"id":"2511.23174","title":"Are LLMs Good Safety Agents or a Propaganda Engine?","created":"2025-11-28","cited_by_count":0,"area":"safety"},{"id":"2601.10566","title":"Representation-Aware Unlearning via Activation Signatures: From Suppression to Entity-Signature Erasure","created":"2026-01-15","cited_by_count":0,"area":"safety"},{"id":"2602.20102","title":"BarrierSteer: LLM Safety via Learning Barrier Steering","created":"2026-02-23","cited_by_count":0,"area":"safety"},{"id":"2604.10326","title":"Jailbreaking the Matrix: Nullspace Steering for Controlled Model Subversion","created":"2026-04-11","cited_by_count":0,"area":"safety"},{"id":"2604.21571","title":"Separable Expert Architecture: Toward Privacy-Preserving LLM Personalization via Composable Adapters and Deletable User Proxies","created":"2026-04-23","cited_by_count":0,"area":"safety"},{"id":"2605.03095","title":"Revisiting JBShield: Breaking and Rebuilding Representation-Level Jailbreak Defenses","created":"2026-05-04","cited_by_count":0,"area":"safety"},{"id":"2605.05687","title":"DataDignity: Training Data Attribution for Large Language Models","created":"2026-05-07","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":"2606.03486","title":"NeuroArmor: Safe-Variant-Guided Representation Consistency for Selective Re-Anchoring in Jailbreak Defense","created":"2026-06-02","cited_by_count":0,"area":"safety"},{"id":"2606.05958","title":"Steering Vectors are an Adversarial Attack Surface","created":"2026-06-04","cited_by_count":0,"area":"safety"},{"id":"2606.09388","title":"Distilling Safe LLM Systems via Soft Prompts for On Device Settings","created":"2026-06-08","cited_by_count":0,"area":"safety"},{"id":"2606.28525","title":"A Gravitational Interpretation of Safety Reversion under Fine-Tuning","created":"2026-06-26","cited_by_count":0,"area":"safety"},{"id":"2607.22595","title":"xMIx: High-Performance Serving-Time Platform for Mechanistic Interpretability Apps","created":"2026-06-11","cited_by_count":0,"area":"safety"},{"id":"2608.08212","title":"Harmful Content Is Not Enough: Continuation Framing Moderates In-Context Emergent Misalignment","created":"2026-08-08","cited_by_count":0,"area":"safety"},{"id":"2608.12652","title":"Excess Separability: Nuisance-Controlled Residual-Stream Probing for Benchmark Contamination Detection","created":"2026-08-12","cited_by_count":0,"area":"safety"},{"id":"2608.18089","title":"Latent Space Refusal Anchoring for Low-Resource African Languages: Mechanistic Safety Recovery Without Retraining","created":"2026-06-05","cited_by_count":0,"area":"safety"},{"id":"2608.23873","title":"Semantic Overlays: Mitigating Prompt Injection with Annotations Beyond Tokens and Steering Vectors","created":"2026-08-24","cited_by_count":0,"area":"safety"},{"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":"2608.30585","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":"2609.00790","title":"RISA: Response Inspection and Selective Actions for Refusal Calibration in Large Language Models","created":"2026-09-01","cited_by_count":0,"area":"safety"},{"id":"2609.13478","title":"BadEngram: Backdoor Attack on Gated Memory Components in LLMs","created":"2026-09-11","cited_by_count":0,"area":"safety"},{"id":"2609.16229","title":"Test-Time Unlearning via Sparse Autoencoder","created":"2026-09-14","cited_by_count":0,"area":"safety"},{"id":"2609.21996","title":"A Lie Detector Test for Language Models: Reading Knowledge a Model Won't Reveal","created":"2026-09-18","cited_by_count":0,"area":"safety"},{"id":"2609.34514","title":"How to Tame a Multi-Headed Hydra? Adaptive Multi-Category Safety Steering for Large Language Models","created":"2026-09-28","cited_by_count":0,"area":"safety"},{"id":"2609.35544","title":"Less Sycophancy, Stronger Refusal? Lessons for AI Safety from Mechanistic Interpretability","created":"2026-09-28","cited_by_count":0,"area":"safety"},{"id":"2609.35913","title":"UNBIND: UNlearning By INference-time Directional Steering for Code LLMs","created":"2026-09-28","cited_by_count":0,"area":"safety"},{"id":"2609.36657","title":"Constitutional adapters: Inference-time interventions for misalignment and misuse","created":"2026-09-29","cited_by_count":0,"area":"safety"},{"id":"2609.39573","title":"Steering Fields: Adaptive Vector Fields for Safe Image Generation and Beyond","created":"2026-09-30","cited_by_count":0,"area":"safety"},{"id":"2609.39882","title":"LLM Persona Unlearning","created":"2026-09-30","cited_by_count":0,"area":"safety"},{"id":"2610.00348","title":"Removing the NEEDLE in the Haystack: Backdoor Removal in LLMs via Weight Orthogonalisation","created":"2026-09-29","cited_by_count":0,"area":"safety"},{"id":"2610.04245","title":"On the Steering Dimensionality of Refusal in Language Models","created":"2026-10-03","cited_by_count":0,"area":"safety"},{"id":"2610.06064","title":"TrustMI: Causally controlling how assistants trust their users","created":"2026-10-05","cited_by_count":0,"area":"safety"},{"id":"2610.15218","title":"Running Fewer Loops Removes Refusal Behaviour in Recurrent-Depth Language Models","created":"2026-10-10","cited_by_count":0,"area":"safety"},{"id":"2610.17810","title":"Perturbation, Not Provenance: Why Activation Steering Does Not Make a Robust Text Watermark","created":"2026-10-13","cited_by_count":0,"area":"safety"},{"id":"2610.17880","title":"Cross-Lingual Transfer of Evil Vectors Is Directionally Asymmetric","created":"2026-10-13","cited_by_count":0,"area":"safety"},{"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}