{"total":191,"rows":[{"id":"2608.02657","title":"Your Agentic LLMs Secretly Encode Indirect Prompt-Injection Exposure in Hidden States","created":"2026-08-01","cited_by_count":7,"area":"agents"},{"id":"2606.03002","title":"Perplexity Can Miss SAE Feature Damage Under Quantization","created":"2026-06-02","cited_by_count":5,"area":"mechinterp"},{"id":"2606.11722","title":"ICA Lens: Interpreting Language Models Without Training Another Dictionary","created":"2026-06-10","cited_by_count":4,"area":null},{"id":"2606.11893","title":"Beyond representational alignment with brain-guided language models for robust reasoning","created":"2026-06-10","cited_by_count":4,"area":"lr"},{"id":"2606.20097","title":"HydraHead: From Head-Level Functional Heterogeneity to Specialized Attention Hybridization","created":"2026-06-18","cited_by_count":4,"area":null},{"id":"2605.31183","title":"Steering LLMs? Actually, Sparse Autoencoders can outperform simple baselines","created":"2026-05-29","cited_by_count":3,"area":"mechinterp"},{"id":"2606.20678","title":"Beyond Importance: Interchange-Sobol Sensitivity Reveals Task-Specific Content Channels in Transformer Components","created":"2026-06-12","cited_by_count":3,"area":"mechinterp"},{"id":"2606.29604","title":"Mechanistically Eliciting Latent Behaviors in Language Models","created":"2026-06-28","cited_by_count":3,"area":null},{"id":"2607.12624","title":"PVDetector: Detecting Prompt Injection Attacks on Purpose-Specific LLM Agents through Policy-Violation Concept Analysis","created":"2026-07-14","cited_by_count":3,"area":"agents"},{"id":"2608.13474","title":"Decoding Task Progress from VLA Representations","created":"2026-08-13","cited_by_count":3,"area":"vla"},{"id":"2609.07876","title":"LLM Layers Immediately Correct Each Other","created":"2026-09-07","cited_by_count":3,"area":"mechinterp"},{"id":"2605.31518","title":"On the Relationship Between Activation Outliers and Feature Death in Sparse Autoencoders","created":"2026-05-29","cited_by_count":2,"area":"mechinterp"},{"id":"2606.01189","title":"The Case for Model Science: Verify, Explore, Steer, Refine","created":"2026-05-31","cited_by_count":2,"area":null},{"id":"2606.05378","title":"Pattern Selectivity is Not Task-Causal Structure: A Cross-Architecture Mechanistic Study of Composed-Task Circuits in 1B-Class Language Models","created":"2026-06-03","cited_by_count":2,"area":"moe"},{"id":"2606.14805","title":"Knowledge-Based Zero-Replay Debugging of Multi-Agent LLM Traces","created":"2026-06-11","cited_by_count":2,"area":"agents"},{"id":"2606.23443","title":"What Does a Chemical Language Model Know About Molecules?","created":"2026-06-22","cited_by_count":2,"area":null},{"id":"2606.24716","title":"Evaluating the Interpretability of Sparse Autoencoders with Concept Annotations","created":"2026-06-23","cited_by_count":2,"area":"mllm"},{"id":"2606.26228","title":"Interpreting \"Interpretability\" and Explaining \"Explainability\" in Machine Learning in Physics","created":"2026-06-24","cited_by_count":2,"area":null},{"id":"2606.26629","title":"From Weights to Features: SAE-Guided Activation Regularization for LLM Continual Learning","created":"2026-06-25","cited_by_count":2,"area":"mechinterp"},{"id":"2606.30449","title":"Internal-State Probes Read the Situation, Not the Action: Three Negative Results for Pre-Action Misalignment Monitoring","created":"2026-06-29","cited_by_count":2,"area":"agents"},{"id":"2606.32022","title":"SemRF: A Semantic Reference Frame for Residual-Stream Dynamics in Language Models","created":"2026-06-30","cited_by_count":2,"area":null},{"id":"2607.04432","title":"Covert Trait Propagation Is Representation Alignment: Mechanistic Evidence from Hidden-Channel Distillation","created":"2026-07-05","cited_by_count":2,"area":null},{"id":"2607.27617","title":"Hidden APIs in Language Models: Discovering Reusable Causal Interfaces from Forked Futures","created":"2026-07-30","cited_by_count":2,"area":null},{"id":"2608.08159","title":"When Is a Steerable Concept Representation Real? Measurement Confounds in a Cross-Family Audit of Neuroscience Parallels in LLMs","created":"2026-08-08","cited_by_count":2,"area":null},{"id":"2606.01695","title":"CANARY: Zero-Label Detection of Fine-Tuning Contamination in Language Models","created":"2026-06-01","cited_by_count":1,"area":"mechinterp"},{"id":"2606.02528","title":"Auditing Asset-Specific Preferences in Financial Large Language Models: Evidence from Bitcoin Representations and Portfolio Allocation","created":"2026-06-01","cited_by_count":1,"area":null},{"id":"2606.02765","title":"Representational Capacity: Geometric Limits on Feature Representation in Transformer Language Models","created":"2026-06-01","cited_by_count":1,"area":null},{"id":"2606.07007","title":"A Geometric View for Understanding Concept Learning and Neuron Interpretation in Sparse Autoencoders","created":"2026-06-05","cited_by_count":1,"area":"mechinterp"},{"id":"2606.07714","title":"Beyond Accuracy: Interpreting Topic Representation in Suicide Ideation Detection Models","created":"2026-06-05","cited_by_count":1,"area":null},{"id":"2606.10080","title":"VFUSE: Virulent Feature Understanding with Sparse autoEncoders","created":"2026-06-08","cited_by_count":1,"area":"mechinterp"},{"id":"2606.15092","title":"High-Dimensional Random Projection for Activation Steering in Language Models","created":"2026-06-13","cited_by_count":1,"area":"mechinterp"},{"id":"2606.15420","title":"Constitutional Value Potentials: reading and steering internal priority margins in language models","created":"2026-06-13","cited_by_count":1,"area":null},{"id":"2606.16193","title":"Cascaded Sparse Autoencoders Learn Multi-Level Visual Concepts in Multimodal LLMs","created":"2026-06-15","cited_by_count":1,"area":"mllm"},{"id":"2606.18322","title":"SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior","created":"2026-06-16","cited_by_count":1,"area":"mechinterp"},{"id":"2606.24964","title":"Evidence for feature-specific error correction in LLMs","created":"2026-06-23","cited_by_count":1,"area":"mechinterp"},{"id":"2606.26474","title":"Localizing RL-Induced Tool Use to a Single Crosscoder Feature","created":"2026-06-25","cited_by_count":1,"area":"agents"},{"id":"2606.28548","title":"Turn-Averaged SAEs for Feature Discovery and Long-Context Attribution","created":"2026-06-26","cited_by_count":1,"area":"mechinterp"},{"id":"2607.01047","title":"Conversable Complexity: Agentic LLM Collectives as Interpretable Substrates","created":"2026-07-01","cited_by_count":1,"area":"agents"},{"id":"2607.01336","title":"Mechanistic Interpretability and Causal Feature Steering of Neural Quantum States via Sparse Autoencoders","created":"2026-07-01","cited_by_count":1,"area":"mechinterp"},{"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":"2607.03372","title":"Present but Not Remembered: Auditing How Frozen VLAs Encode, Deploy, and Steer Visual History","created":"2026-07-03","cited_by_count":1,"area":null},{"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":"2607.08946","title":"Training, Reading, and Editing Legible Transformers","created":"2026-07-09","cited_by_count":1,"area":null},{"id":"2607.13346","title":"The Refusal Residue: When Probes Catch Alignment Faking and When They Don't","created":"2026-07-15","cited_by_count":1,"area":null},{"id":"2607.16448","title":"Retrieval is Enough: Training-Free Interpretability with a Tool-Using Agent","created":"2026-07-17","cited_by_count":1,"area":"agents"},{"id":"2607.18454","title":"Estimating Rare Events in Language Models with Proper Evaluation","created":"2026-07-20","cited_by_count":1,"area":null},{"id":"2607.20058","title":"Reading and Steering Representations of Materials-Science Mechanisms in an Open-Weight Language Model","created":"2026-07-22","cited_by_count":1,"area":null},{"id":"2607.22570","title":"Reference Feature Atlases for Mechanistic Auditing of Language Models","created":"2026-06-01","cited_by_count":1,"area":null},{"id":"2607.22676","title":"How LLM Task-Adaptation Reshapes Alignment: A Multi-dimensional Study of Behavioral and Representational Drift","created":"2026-07-10","cited_by_count":1,"area":null},{"id":"2607.24645","title":"Sparse Autoencoders Encode Both Concepts and Functions: The Downstream Geometry of Feature Effects","created":"2026-07-27","cited_by_count":1,"area":"mechinterp"},{"id":"2608.19211","title":"Represented but Ignored: A Causal Account of Prosodic Underuse in Audio-Language Models","created":"2026-06-12","cited_by_count":1,"area":null},{"id":"2609.07037","title":"Disentangling Steering Vectors","created":"2026-09-07","cited_by_count":1,"area":"mechinterp"},{"id":"2609.32407","title":"Opening LLM Judges: Recovering Preference Signals Beyond the Final Verdict","created":"2026-09-26","cited_by_count":1,"area":null},{"id":"2605.31304","title":"Interpretability Without Tradeoffs: Disentangling Polysemanticity At Equal Predictive Performance","created":"2026-05-29","cited_by_count":0,"area":null},{"id":"2606.00566","title":"Same Payload, Different Channel: Measuring Trust Asymmetry in Tool-Using Language Models","created":"2026-05-30","cited_by_count":0,"area":"agents"},{"id":"2606.00603","title":"Toward Agentic Governance: What Shapes LLM-Agent Intervention in Public Forums?","created":"2026-05-30","cited_by_count":0,"area":"agents"},{"id":"2606.00930","title":"A Circuit, Not The Circuit: Non-Unique Causal Localisation of the Mamba-2 State Sink","created":"2026-05-30","cited_by_count":0,"area":"ssm"},{"id":"2606.02385","title":"How Optimality Structures Sparse Dictionaries: Theory for Interpreting SAE Representations","created":"2026-06-01","cited_by_count":0,"area":"mechinterp"},{"id":"2606.04160","title":"Decomposing Refusal Steering in Mixture-of-Experts Models","created":"2026-06-02","cited_by_count":0,"area":"alignment"},{"id":"2606.06315","title":"LLM Self-Recognition: Steering and Retrieving Activation Signatures","created":"2026-06-04","cited_by_count":0,"area":null},{"id":"2606.07617","title":"Query Lens: Interpreting Sparse Key-Value Features with Indirect Effects","created":"2026-05-30","cited_by_count":0,"area":"mechinterp"},{"id":"2606.08296","title":"Revisiting the shutdown problem","created":"2026-06-06","cited_by_count":0,"area":null},{"id":"2606.08365","title":"Pre-Intervention Prediction of Sparse Autoencoder Steering Side Effects","created":"2026-06-06","cited_by_count":0,"area":"mechinterp"},{"id":"2606.08832","title":"Instrumental convergence and power-seeking","created":"2026-06-07","cited_by_count":0,"area":null},{"id":"2606.09940","title":"Interactions Between Crosscoder Features: A Compact Proofs Perspective","created":"2026-06-08","cited_by_count":0,"area":"mechinterp"},{"id":"2606.10669","title":"In Defense of Information Leakage in Concept-based Models","created":"2026-06-09","cited_by_count":0,"area":null},{"id":"2606.10747","title":"The Arbiter Agent: Continually Monitoring Multi-Agent Conversations to Detect Emergent Misalignment","created":"2026-06-09","cited_by_count":0,"area":"agents"},{"id":"2606.12629","title":"The Signs Were Always There: Training-Free Concept Detection and Steering in Raw Transformer Dimensions","created":"2026-06-10","cited_by_count":0,"area":null},{"id":"2606.14040","title":"Decompose Sparsely Where You Should, Absorb Densely Where You Should No","created":"2026-06-12","cited_by_count":0,"area":"mechinterp"},{"id":"2606.15468","title":"Analyzing Visual Aircraft Representations with Sparse Autoencoders","created":"2026-06-13","cited_by_count":0,"area":"mechinterp"},{"id":"2606.16044","title":"Circuit Tracing in Autoregressive Protein Language Models","created":"2026-06-14","cited_by_count":0,"area":"mechinterp"},{"id":"2606.16240","title":"Creative Collision: Directorial Persona Steering and Competition in Large Language Models","created":"2026-06-15","cited_by_count":0,"area":"mechinterp"},{"id":"2606.17010","title":"From Tokens to Policy: Causal and Interpretable Heterogeneous Treatment Effects Identification","created":"2026-06-15","cited_by_count":0,"area":null},{"id":"2606.20347","title":"Critical Percolation as a Synthetic Data Model for Interpretability","created":"2026-06-18","cited_by_count":0,"area":null},{"id":"2606.21195","title":"Beyond Hooking Onto the World: Referential Profiles and the Numerical Structure of LLM Grounding","created":"2026-06-19","cited_by_count":0,"area":null},{"id":"2606.26396","title":"At the Edge of Understanding: Sparse Autoencoders Trace The Limits of Transformer Generalization","created":"2026-06-24","cited_by_count":0,"area":"mechinterp"},{"id":"2606.27440","title":"PairSAE: Mechanistic Interpretability from Pair Representations in Protein Co-Folding","created":"2026-06-25","cited_by_count":0,"area":"mechinterp"},{"id":"2606.28770","title":"Mechanistic Personality Analysis of LLMs: Steering Personality via Latent Feature Interventions","created":"2026-06-27","cited_by_count":0,"area":"mechinterp"},{"id":"2606.29888","title":"Same Concept, Different Directions: Cross-Modal Feature Heterogeneity in Sparse Autoencoders","created":"2026-06-29","cited_by_count":0,"area":"mllm"},{"id":"2606.31394","title":"Resolving superposition in AI for interpretability and cross-modal alignment in patient-neuronal images","created":"2026-06-30","cited_by_count":0,"area":null},{"id":"2607.00601","title":"\"Don't Say It!\": Constraints, Compliance, and Communication when Language Models Play Taboo","created":"2026-07-01","cited_by_count":0,"area":null},{"id":"2607.01799","title":"Expander Sparse Autoencoders: Parameter-Efficient Dictionaries for Mechanistic Interpretability","created":"2026-07-02","cited_by_count":0,"area":"mechinterp"},{"id":"2607.03365","title":"Brand-as-Memory: Vision-Language Models Encode Causal, Mechanistically Localizable Credibility Priors for News Sources","created":"2026-07-03","cited_by_count":0,"area":"mllm"},{"id":"2607.04167","title":"Geometry of Ordinal Representations in Language Models","created":"2026-07-05","cited_by_count":0,"area":null},{"id":"2607.05407","title":"Position: Preventing AI-Generated CSAM Necessitates New Approaches to AI Safety","created":"2026-06-09","cited_by_count":0,"area":null},{"id":"2607.06328","title":"Driving the Wrong Way: Leveraging Interpretability in End2End Autonomous Driving Models","created":"2026-07-07","cited_by_count":0,"area":"driving"},{"id":"2607.10578","title":"Laguerre Geometry for Interpreting Large Language Models","created":"2026-07-12","cited_by_count":0,"area":null},{"id":"2607.12166","title":"From Geometric Recovery to Causal Validation: A Reproducible Audit of Sparse Autoencoder Features, from Superposition Geometry to Causal Inertness","created":"2026-07-13","cited_by_count":0,"area":"mechinterp"},{"id":"2607.13899","title":"AIMO Interpretability Challenge","created":"2026-07-15","cited_by_count":0,"area":"lr"},{"id":"2607.15893","title":"Induction in Both Directions: A Mechanistic Analysis of In-Context Learning in Masked Diffusion Language Models","created":"2026-07-17","cited_by_count":0,"area":"diffusion"},{"id":"2607.16295","title":"Emergent Hierarchical Monosemantic Neurons from the Group-Contrastive Forward-Forward Algorithm","created":"2026-07-13","cited_by_count":0,"area":"mechinterp"},{"id":"2607.17117","title":"Persistent Sparse Autoencoders: Learning Feature-Specific Timescales in Language Model Representations","created":"2026-07-19","cited_by_count":0,"area":"mechinterp"},{"id":"2607.18691","title":"Semantic Primes as Explanans for Emotion in Large Language Models","created":"2026-07-21","cited_by_count":0,"area":null},{"id":"2607.19360","title":"Lifted State Hypothesis in Large Language Models","created":"2026-06-02","cited_by_count":0,"area":null},{"id":"2607.19618","title":"Causal dictionary learning reveals and validates transcription-factor binding features in genomic language models","created":"2026-07-21","cited_by_count":0,"area":null},{"id":"2607.20652","title":"Scaling Interpretable Transformers with Parity Bottleneck Layers","created":"2026-07-22","cited_by_count":0,"area":"mechinterp"},{"id":"2607.21491","title":"What, Where, and How: Disentangling the Roles of Task, Language, and Model in Code Model Representations","created":"2026-07-23","cited_by_count":0,"area":null},{"id":"2607.25244","title":"CADENCE: A Cardiac Atom Dictionary for Interpretable Neural Concept Extraction from ECG Foundation Models","created":"2026-07-28","cited_by_count":0,"area":null},{"id":"2607.26825","title":"From Found to Designed: Concepts as a Design Axis for Large Language Models","created":"2026-07-29","cited_by_count":0,"area":null},{"id":"2607.26853","title":"From Representations to Behaviors: Exploring the Person-Situation-Behavior Triad in LLMs","created":"2026-07-29","cited_by_count":0,"area":null}],"next_cursor":"100"}