{"total":83,"rows":[{"id":"2505.05410","title":"Reasoning Models Don't Always Say What They Think","created":"2025-05-08","cited_by_count":415,"area":"lr"},{"id":"2511.18397","title":"Natural Emergent Misalignment from Reward Hacking in Production RL","created":"2025-11-23","cited_by_count":136,"area":"alignment"},{"id":"2607.15495","title":"Verbalizable Representations Form a Global Workspace in Language Models","created":"2026-07-16","cited_by_count":75,"area":null},{"id":"2604.13602","title":"Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges","created":"2026-04-15","cited_by_count":33,"area":"alignment"},{"id":"2509.00328","title":"Mechanistic interpretability for steering vision-language-action models","created":"2025-08-30","cited_by_count":31,"area":"vla"},{"id":"2506.14261","title":"RL-Obfuscation: Can Language Models Learn to Evade Latent-Space Monitors?","created":"2025-06-17","cited_by_count":24,"area":null},{"id":"2605.02087","title":"Model Spec Midtraining: Improving How Alignment Training Generalizes","created":"2026-05-03","cited_by_count":20,"area":null},{"id":"2505.14352","title":"Towards eliciting latent knowledge from LLMs with mechanistic interpretability","created":"2025-05-20","cited_by_count":19,"area":"mechinterp"},{"id":"2506.12152","title":"Because we have LLMs, we Can and Should Pursue Agentic Interpretability","created":"2025-06-13","cited_by_count":16,"area":"agents"},{"id":"2505.13778","title":"CoIn: Counting the Invisible Reasoning Tokens in Commercial Opaque LLM APIs","created":"2025-05-19","cited_by_count":14,"area":"lr"},{"id":"2604.11806","title":"Detecting Safety Violations Across Many Agent Traces","created":"2026-04-13","cited_by_count":14,"area":null},{"id":"2511.22662","title":"Difficulties with Evaluating a Deception Detector for AIs","created":"2025-11-27","cited_by_count":13,"area":null},{"id":"2604.16812","title":"Introspection Adapters: Training LLMs to Report Their Learned Behaviors","created":"2026-04-18","cited_by_count":11,"area":null},{"id":"2506.20702","title":"The Singapore Consensus on Global AI Safety Research Priorities","created":"2025-06-25","cited_by_count":10,"area":null},{"id":"2507.08218","title":"Simple Mechanistic Explanations for Out-Of-Context Reasoning","created":"2025-07-10","cited_by_count":10,"area":"lr"},{"id":"2510.27338","title":"Reasoning Models Sometimes Output Illegible Chains of Thought","created":"2025-10-31","cited_by_count":9,"area":"lr"},{"id":"2603.03992","title":"Measuring AI R&D Automation","created":"2026-03-04","cited_by_count":9,"area":null},{"id":"2512.22154","title":"Practical challenges of control monitoring in frontier AI deployments","created":"2025-12-15","cited_by_count":8,"area":null},{"id":"2602.16698","title":"Causality is Key for Interpretability Claims to Generalise","created":"2026-02-18","cited_by_count":8,"area":null},{"id":"2603.22582","title":"Lie to Me: How Faithful Is Chain-of-Thought Reasoning in Reasoning Models?","created":"2026-03-23","cited_by_count":8,"area":"lr"},{"id":"2505.19690","title":"Beyond Safe Answers: A Benchmark for Evaluating True Risk Awareness in Large Reasoning Models","created":"2025-05-26","cited_by_count":7,"area":"lr"},{"id":"2509.25302","title":"Dive into the Agent Matrix: A Realistic Evaluation of Self-Replication Risk in LLM Agents","created":"2025-09-29","cited_by_count":7,"area":"agents"},{"id":"2606.22019","title":"Channel Location Constrains the Auditability of Subliminal Learning","created":"2026-06-20","cited_by_count":7,"area":null},{"id":"2607.01033","title":"The Model Organism Lottery: Model Organism Interpretability Strongly Depends on Training Methodology","created":"2026-07-01","cited_by_count":6,"area":"mechinterp"},{"id":"2602.02581","title":"QuantLRM: Quantization of Large Reasoning Models via Fine-Tuning Signals","created":"2026-01-31","cited_by_count":5,"area":"lr"},{"id":"2605.10310","title":"Positive Alignment: Artificial Intelligence for Human Flourishing","created":"2026-05-11","cited_by_count":5,"area":null},{"id":"2607.22925","title":"Not All LLM Reasoning is Visible in the Chain-of-Thought","created":"2026-07-24","cited_by_count":5,"area":"lr"},{"id":"2506.10138","title":"Path Channels and Plan Extension Kernels: a Mechanistic Description of Planning in a Sokoban RNN","created":"2025-06-11","cited_by_count":4,"area":"deeprl"},{"id":"2508.09218","title":"Towards Effective MLLM Jailbreaking Through Balanced On-Topicness and OOD-Intensity","created":"2025-08-11","cited_by_count":4,"area":"mllm"},{"id":"2603.20172","title":"Measuring Faithfulness Depends on How You Measure: Classifier Sensitivity in LLM Chain-of-Thought Evaluation","created":"2026-03-20","cited_by_count":4,"area":"lr"},{"id":"2604.11061","title":"Pando: Do Interpretability Methods Work When Models Won't Explain Themselves?","created":"2026-04-13","cited_by_count":4,"area":"mechinterp"},{"id":"2604.13076","title":"Alignment midtraining for animals","created":"2026-03-21","cited_by_count":4,"area":null},{"id":"2605.28591","title":"Models That Know How Evaluations Are Designed Score Safer","created":"2026-05-27","cited_by_count":4,"area":null},{"id":"2605.30322","title":"Gram: Assessing sabotage propensities via automated alignment auditing","created":"2026-05-28","cited_by_count":4,"area":"agents"},{"id":"2510.16255","title":"Detecting Adversarial Fine-tuning with Auditing Agents","created":"2025-10-17","cited_by_count":3,"area":"agents"},{"id":"2603.26676","title":"Evaluating Human-AI Safety: A Framework for Measuring Harmful Capability Uplift","created":"2026-03-06","cited_by_count":3,"area":null},{"id":"2604.13466","title":"Functional Emotions or Situational Contexts? A Discriminating Test from the Mythos Preview System Card","created":"2026-04-09","cited_by_count":3,"area":"mechinterp"},{"id":"2606.07054","title":"TRACE: Trajectory Reasoning through Adaptive Cross-Step Evidence Aggregation for LLM Agents","created":"2026-06-05","cited_by_count":3,"area":"lr"},{"id":"2606.11063","title":"CIAware-Bench: Benchmarking Control Intervention Awareness Across Frontier LLMs","created":"2026-06-09","cited_by_count":3,"area":null},{"id":"2606.30252","title":"Inoculation Adapters: Improved Selective Generalization of Capabilities with Fewer Surprising Backdoors","created":"2026-06-29","cited_by_count":3,"area":null},{"id":"2607.18966","title":"Measuring Reward-Seeking via Contrastive Belief Updates","created":"2026-07-21","cited_by_count":3,"area":"alignment"},{"id":"2507.21839","title":"Against racing to AGI: Cooperation, deterrence, and catastrophic risks","created":"2025-07-29","cited_by_count":2,"area":null},{"id":"2509.21344","title":"Linear probes rely on textual evidence: Results from leakage mitigation studies in language models","created":"2025-09-16","cited_by_count":2,"area":"mechinterp"},{"id":"2511.00180","title":"ParaScopes: What do Language Models Activations Encode About Future Text?","created":"2025-10-31","cited_by_count":2,"area":null},{"id":"2605.15164","title":"Position: Behavioural Assurance Cannot Verify the Safety Claims Governance Now Demands","created":"2026-05-14","cited_by_count":2,"area":null},{"id":"2609.19101","title":"Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations","created":"2026-09-16","cited_by_count":2,"area":"alignment"},{"id":"2506.22740","title":"Explanations are a Means to an End: Decision Theoretic Explanation Evaluation","created":"2025-06-28","cited_by_count":1,"area":null},{"id":"2602.08964","title":"A Behavioural and Representational Evaluation of Goal-Directedness in Language Model Agents","created":"2026-02-09","cited_by_count":1,"area":"agents"},{"id":"2605.06846","title":"Narrow Secret Loyalty Dodges Black-Box Audits","created":"2026-05-07","cited_by_count":1,"area":"safety"},{"id":"2605.12813","title":"REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations","created":"2026-05-12","cited_by_count":1,"area":null},{"id":"2606.13310","title":"RogueAI: A Reverse Turing Test for Detecting Licensed AI Deception in Dialogue","created":"2026-06-11","cited_by_count":1,"area":null},{"id":"2606.18327","title":"Self-CTRL: Self-Consistency Training with Reinforcement Learning","created":"2026-06-16","cited_by_count":1,"area":"lr"},{"id":"2606.26071","title":"Model Forensics: Investigating Whether Concerning Behavior Reflects Misalignment","created":"2026-06-24","cited_by_count":1,"area":null},{"id":"2607.21518","title":"Same Dangerous Objective, Opposite Advice: Direct Exposure versus Multi-Agent Mediation","created":"2026-07-23","cited_by_count":1,"area":"agents"},{"id":"2607.22570","title":"Reference Feature Atlases for Mechanistic Auditing of Language Models","created":"2026-06-01","cited_by_count":1,"area":null},{"id":"2607.26173","title":"Shared SFT Lessons Across Alignment, Model Organisms, and Toy Models","created":"2026-07-28","cited_by_count":1,"area":null},{"id":"2609.25518","title":"Matryoshka attribution: Learning to attribute language model outputs to representations and weights","created":"2026-09-22","cited_by_count":1,"area":null},{"id":"2506.04774","title":"Fine-Grained Interpretation of Political Opinions in Large Language Models","created":"2025-06-05","cited_by_count":0,"area":"mechinterp"},{"id":"2508.06592","title":"Towards Integrated Alignment","created":"2025-08-08","cited_by_count":0,"area":"alignment"},{"id":"2512.01514","title":"Label Forensics: Interpreting Hard Labels in Black-Box Text Classifier","created":"2025-12-01","cited_by_count":0,"area":null},{"id":"2512.13821","title":"The Double Life of Code World Models: Provably Unmasking Malicious Behavior Through Execution Traces","created":"2025-12-15","cited_by_count":0,"area":"wm"},{"id":"2602.08877","title":"Stress-Testing Alignment Audits With Prompt-Level Strategic Deception","created":"2026-02-09","cited_by_count":0,"area":"safety"},{"id":"2602.15338","title":"Discovering Implicit Large Language Model Alignment Objectives","created":"2026-02-17","cited_by_count":0,"area":"alignment"},{"id":"2603.26846","title":"Stable Reasoning, Unstable Responses: Mitigating LLM Deception via Stability Asymmetry","created":"2026-03-27","cited_by_count":0,"area":"lr"},{"id":"2605.00994","title":"Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives","created":"2026-05-01","cited_by_count":0,"area":null},{"id":"2605.11448","title":"Deep Minds and Shallow Probes","created":"2026-05-12","cited_by_count":0,"area":null},{"id":"2607.03478","title":"Demonstrating Generalization Failures via Mixtures of Conditional Policies","created":"2026-07-03","cited_by_count":0,"area":null},{"id":"2608.13840","title":"ASSERT: A Measurement Pipeline for GenAI Audits","created":"2026-08-14","cited_by_count":0,"area":null},{"id":"2608.17524","title":"Evaluating RL Explainability Methods by How Much They Help Fix Bugs in Agents","created":"2026-08-18","cited_by_count":0,"area":"agents"},{"id":"2608.25460","title":"Training Alignment Auditors via Reinforcement Learning","created":"2026-08-26","cited_by_count":0,"area":null},{"id":"2609.10883","title":"Story Imprinting: AI Assistants Absorb Traits from Human Characters They Resemble","created":"2026-09-09","cited_by_count":0,"area":null},{"id":"2609.30266","title":"LLM Agents Can Easily Tamper With Their Own Traces","created":"2026-09-24","cited_by_count":0,"area":"agents"},{"id":"2609.31186","title":"Evolutionary Safety of Recursive Self-Improving AI: Taxonomy, Risk Discovery, and Evaluation","created":"2026-09-25","cited_by_count":0,"area":null},{"id":"2609.32361","title":"Black-Box Auditing of Epistemic Reliability in Multi-Agent Debate Distillation","created":"2026-09-26","cited_by_count":0,"area":"agents"},{"id":"2610.00767","title":"Pre-training interventions, ex post facto: Grafting model beliefs across checkpoints","created":"2026-09-30","cited_by_count":0,"area":null},{"id":"2610.04441","title":"LocusRL: Diagnosing LLM Reward and Policy Interventions in Competitive Games","created":"2026-10-03","cited_by_count":0,"area":null},{"id":"2610.05793","title":"Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent System","created":"2026-10-05","cited_by_count":0,"area":"agents"},{"id":"2610.07518","title":"Harmful SFT Leaves a Continuous Trace in LLM Checkpoint Updates","created":"2026-10-05","cited_by_count":0,"area":null},{"id":"2610.10203","title":"How to train your model organism","created":"2026-10-07","cited_by_count":0,"area":"safety"},{"id":"2610.11012","title":"Distillation for Incrimination and Distillation for Capabilities","created":"2026-10-07","cited_by_count":0,"area":null},{"id":"2610.15365","title":"TRACE: Temporal Risk Assessment and Concept Editing for Safer Language Generation","created":"2026-10-11","cited_by_count":0,"area":"safety"},{"id":"2610.16911","title":"Audit tools of the trade: Automated Auditor Elicitation Can Be Sensitive to the Context the Auditor Is Given","created":"2026-10-12","cited_by_count":0,"area":"alignment"},{"id":"2610.18727","title":"On the Monitorability of Natural Language Autoencoders: Identification vs Discrimination","created":"2026-10-13","cited_by_count":0,"area":null}],"next_cursor":null}